DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This is a Final Office Action in response to the amendment filed on March 17, 2026, for application 17/554,837 entitled "SYSTEM AND METHOD FOR PROCESSING INSURANCE CARD”.
Status of Claims
Claims 1, 4, and 5 have been amended and are hereby entered.
Claims 8, 10-18 were previously cancelled.
Claim 20 is added.
Claims 1-7, 9, and 19-20 are pending and have been examined.
Response to Amendment
The amendment filed March 17, 2026, has been entered. Claims 1-7, 9, and 19-20 remain pending in the application. Applicant’s amendments to the Specification, Drawings, and/or Claims have been noted in response to the Non-Final Office Action mailed December 29, 2025.
Examiner’s Note
Intended use language is generally given limited patentable weight. See MPEP 2114(II) ("A claim containing a 'recitation with respect to the manner in which a claimed apparatus is intended to be employed does not differentiate the claimed apparatus from a prior art apparatus’ if the prior art apparatus teaches all the structural limitations of the claim. Ex parte Masham, 2 USPQ2d 1647 (Bd. Pat. App. & Inter. 1987).”); see also MPEP 2103(C). Examples of claim limitations that are often found to precede intended use include “adapted to,” “capable of,” “sufficient to,” “whereby,” and “for.” The following limitations are interpreted as intended use limitations:
The examiner would like to note that the claims are replete with intended use, however, to provide compact prosecution, the examiner has provided the mapping and rejections.
Claim 1: “to identify characters on the insurance card”, “to correspond to an information item of interest”, “to improve subsequent extraction accuracy”
Claims 3 and 5: “to reflect physical spacing on the card”
Claim 20: “… for processing the tokens as character sequences along with spatial coordinates… for integrating context from neighboring tokens based on spatial proximity, … for processing the image of the insurance card ….”
The Examiner determines the aforementioned intended use statements do not result in any structural nor manipulative difference between the claimed invention and the prior art. Therefore, the intended use statements are afforded limited patentable weight.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-7, 9, and 19-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Please see MPEP 2106 for additional information regarding Patent Subject Matter Eligibility Guidance.
Claims 1-7, 9, and 19-20 are directed to a system, method/process, machine/apparatus, or composition of matter, which are/is one of the statutory categories of invention. (Step 1: YES).
The claimed invention is directed to an abstract idea without significantly more.
Independent Claim 1 recites:
“A system for processing image data generated from a plurality of images of insurance cards to extract discrete entities of information, said system …to:
receive and process the image data generated from the images of the insurance cards …to identify characters on the insurance cards;
determine a distance between the characters on the insurance cards;
identify a plurality of different subsets of characters on the insurance cards based, at least in part, on said distance between the characters such that each of the subsets include only the characters within a predetermined distance of one another;
identify relative spatial orientation of said characters of each of said subsets to determine a plurality of tokens, one for each of the subsets of characters and a spatial orientation of said tokens, where the tokens each represent one of a plurality of possible combinations of said characters on the insurance card perceived to correspond to an information item of interest comprising a member ID and a payer name for a respective one of the insurance cards such that each of said tokens represents a discrete combination of some or all of the characters representing one of:
a plausible member ID and a plausible payer name for the respective one of the insurance cards, where said plurality of tokens are each modeled as a respective sequence of characters from a set of possible characters including all letters of the alphabet and numbers 0 through 9, and where said plurality of tokens comprise pixel values of the image data;
determine coordinates for each token on the respective one of the insurance cards based on the spatial orientation of the tokens, the tokens and coordinates representing an…output;
…including a node for each token based on the image data generated from the insurance cards and the … output;
scoring, …each node with a member ID score for the likelihood that said node corresponds to the member ID for the respective one of the insurance cards;
identifying, for each of the insurance cards, a member ID for said respective one of the insurance cards based on the node with the highest member ID score and emitting the token associated with the highest member ID score as the member ID;
scoring, …each node with a payer name score for the likelihood that said node corresponds to the payer name for the respective one of the insurance cards;
identifying, for each of the insurance cards, a payer name for said respective one of the insurance cards based on the node with the highest payer name score, and emitting the token associated with the highest payer name score as the payer name;
ranking labels …of predetermined payer labels against the identified payer names;
displaying the identified member ID and a highest ranking one of the
predetermined payer labels; and
storing, for each of the insurance cards. the image data, the …outputs, the identified payer name, and the identified member ID, and…using the stored data
…adjusts parameters …based on feedback from the stored identified payer names and member IDs to improve subsequent extraction accuracy;
where each of said insurance cards comprise words including the payer name and an alphanumeric combination of spaced characters providing the member ID.”
These limitations, under their broadest reasonable interpretation, cover performance of the limitation as certain methods of organizing human activity. Specific instances include instructing to identifying, on each insurance card, a member ID and scoring…for the likelihood that…corresponds to a member ID on the insurance card recite a commercial or legal interactions. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation as a fundamental economic, commercial, or financial action, principle, or practice then it falls within the “Certain Methods of Organizing Human Activity” grouping of abstract ideas. Accordingly, the claim recites an abstract idea. (Step 2A-Prong 1: YES. The claims recite an abstract idea).
This judicial exception is not integrated into a practical application. In particular, the claims recite the additional elements of:
[comprising at least one non-transitory computer-readable medium storing software instructions, which, when executed configure one or more processors] [from a database]:
merely applying computer processing, storage, and networking technology as tools to perform an abstract idea
[extract discrete entities of information]:
insignificant extra-solution activity to the judicial exception of data gathering
[using optical character recognition (OCR)] [OCR] merely applying the generic optical character recognition to the abstract idea.
[generate a fully connected neural network] [by way of said fully connected neural network][by way of said fully connected neural network] [re-training the fully connected network] [wherein said re-training] [of the fully connected neural network]:
merely applying the generic neural networks to the abstract idea.
are recited at a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer components and/or electronic processes. For example, the Applicant’s Specification reads:
[0023] For capturing tokens from the insurance card 102 at OCR module 112, off-the-shelf or in-house optical character recognition systems can be employed.... Several different deep learning architectures can be used, including multimodal architectures which simultaneously process the tokens (character sequences) coming out of the OCR system
[0046] For all the above methods that use neural networks, includes RNNs, CNNs, GNNs, and the like, the training is performed via a gradient descent procedure … can be done through a variety of techniques including but not limited to random normal, Glorot normal, Glorot uniform, and He normal
[0049] While there exist a number of software libraries which assist in the training and deployment of neural networks, one software library that has been found to be advantageous is Tensorflow 2.0 (Google's neural network library), which contains the API needed to construct custom neural networks
[0050] in and out of the deep learning inference pipeline, can be defined as follows….from cloud storage (AWS S3) cloud storage (AWS S3) 314 are loaded into memory using the Tensorflow library
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea and are at a high level of generality. Therefore, Claim 1 is directed to an abstract idea without a practical application. (Step 2A-Prong 2: NO. The additional claimed elements are not integrated into a practical application)
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, the additional elements, do not change the outcome of the analysis, when considered separately and as an ordered combination. The claim further defines the abstract idea and hence is abstract for the reasons presented above. The claim does not include any additional elements that integrate the abstract idea into a practical application or are sufficient to amount to significantly more than the judicial exception when considered both individually and as an ordered combination. Therefore, the claim is directed to an abstract idea. Thus, the claim is not patent eligible. (Step 2B: NO. The claim does not provide significantly more)
Dependent Claims recite additional elements.
This judicial exception is not integrated into a practical application. In particular, the recited additional elements of
Claim 2:
“OCR”: merely applying the generic optical character recognition to the abstract idea
Claim 3:
“OCR”: merely applying the generic optical character recognition to the abstract idea
“fully connected neural network”: merely applying the generic neural networks to the abstract idea
Claim 19:
“one non-transitory computer-readable medium comprises addition software instructions, which, when executed configure said one or more processors to: generate a database”: merely applying computer processing, storage, and networking technology as tools to perform an abstract idea
Claim 20:
“fully connected neural network … a recurrent neural network (RNN) … a graph neural network (GNN) … convolutional neural network (CNN) … from the RNN, and a second RNN modeling the OCR”: merely applying the generic neural networks to the abstract idea
are recited at a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer components and/or electronic processes. For example, the Applicant’s Specification reads:
[0023] For capturing tokens from the insurance card 102 at OCR module 112, off-the-shelf or in-house optical character recognition systems can be employed.... Several different deep learning architectures can be used, including multimodal architectures which simultaneously process the tokens (character sequences) coming out of the OCR system
[0006] a first recurrent neural network (RNN), or RNN variant,... based on a graph neural network (GNN), or GNN variant
[0046] For all the above methods that use neural networks, includes RNNs, CNNs, GNNs, and the like, the training is performed via a gradient descent procedure … can be done through a variety of techniques including but not limited to random normal, Glorot normal, Glorot uniform, and He normal
[0049] While there exist a number of software libraries which assist in the training and deployment of neural networks, one software library that has been found to be advantageous is Tensorflow 2.0 (Google's neural network library), which contains the API needed to construct custom neural networks
[0050] in and out of the deep learning inference pipeline, can be defined as follows….from cloud storage (AWS S3) cloud storage (AWS S3) 314 are loaded into memory using the Tensorflow library
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea and are at a high level of generality. Therefore, the claim is directed to an abstract idea without a practical application. (Step 2A-Prong 2: NO. The additional claimed elements are not integrated into a practical application)
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, these additional elements, do not change the outcome of the analysis, when considered separately and as an ordered combination. Dependent claims further define the abstract idea that is present in their respective independent claims and hence are abstract for the reasons presented above. The dependent claims do not include any additional elements that integrate the abstract idea into a practical application or are sufficient to amount to significantly more than the judicial exception when considered both individually and as an ordered combination. Therefore, the dependent claims are directed to an abstract idea. Thus, the dependent claims are not patent eligible. (Step 2B: NO. The claims do not provide significantly more)
Independent Claim 4 recites:
“A system for processing a plurality of insurance cards, said system comprising:
…configured to capture a plurality of images, the images including at least one image corresponding to each one of the insurance cards, each of said insurance cards each comprising a set of words and an alphanumeric combination of spaced characters providing a member identifier (ID);
…of predetermined payer labels …for each insurance card, to:
receive and process image data generated from the at least one image of the insurance card …to identify characters on the insurance cards;
determine a distance between the characters on the insurance cards;
identify a plurality of different subsets of characters on the insurance card based, at least in part, on said distance between the characters such that each of the subsets include only the characters within a predetermined distance of one another;
identify relative spatial orientation of said characters of each of said subsets to determine a plurality of tokens, one for each of said subsets characters and a spatial orientation of said tokens, where the tokens each represent one of a plurality of possible combinations of said characters on the insurance card perceived to correspond to a member ID such that each of said tokens represents a discrete combination of some or all of the characters corresponding to a plausible member ID for the insurance card, where said plurality of tokens are modeled as a respective sequence of characters from a set of possible characters including all letters of the alphabet and numbers 0 through 9,and where said plurality of tokens comprise pixel values of the image data;
determine coordinates for each token on the insurance card based on the spatial orientation of the tokens;
…to model the …output for each of the plurality of insurance cards, using vector representations of each token to obtain a logit for each token;
…the embeddings being vector representations of the tokens, and using the embeddings and the … output to generate a graph, with each token as a node, to construct a node feature matrix and…;
…processing the image of each of the plurality of insurance cards … to generate an image representation of each insurance card and combining each image representation with a hidden output from the first … wherein the …concatenates, for each of the plurality of insurance cards,
an image embedding of the card with a hidden state …the concatenation forming a fixed length vector input …and wherein …are joined and the parameters of the …are updated simultaneously …
…output using the fixed length vector from the image of the insurance card;
…from each of the plurality of insurance cards based on the processing steps by assigning a score to the tokens based on a likelihood that the token corresponds to the member ID;
and the payer name, respectively;
comparing the discrete entity of information associated with the payer name …of predetermined payer labels and ranking a likelihood of each of the predetermined payer labels:
store said image data, output …and predictive output comprising the payor name and the member ID for …
….using said stored image data,
output from the OCR, and predictive output comprising the payor name and the member ID
…the discrete entity of information associated with the member ID from and the highest ranking one of the predetermined payer labels for each of the plurality of insurance cards to a remote electronic device for display and
generate …information for a plurality of members each
associated with one of the insurance cards, …registering one member for each of the plurality of insurance cards and including the highest ranking one of the predetermined payer labels name and the discrete entity of information associated with the member ID for each registered member”
These limitations, under their broadest reasonable interpretation, cover performance of the limitation as certain methods of organizing human activity. Specific instances include instructing to identifying, on each insurance card, a member ID and scoring…for the likelihood that…corresponds to a member ID on the insurance card recite a commercial or legal interactions. If a claim limitation, under its broadest reasonable interpretation, covers performance of the limitation as a fundamental economic, commercial, or financial action, principle, or practice then it falls within the “Certain Methods of Organizing Human Activity” grouping of abstract ideas. Accordingly, the claim recites an abstract idea. (Step 2A-Prong 1: YES. The claims recite an abstract idea).
This judicial exception is not integrated into a practical application. In particular, the claims recite the additional elements of:
[at least one computer] [electronic device for display or further processing] [at least one computer comprising a database][and software instructions, which when executed, configure the at least onecomputer] [at least one computer comprising a database] [and software instructions, which when executed, configure the at least one computer,] [against the database] [a database of] [the database]:
merely applying computer processing, storage, and networking technology as tools to perform an abstract idea
[a camera]:
merely applying computer imaging technology as tools to perform an abstract idea
[extract at least one discrete entity of information] [ transmit the at least one discrete entity of information] [transmit]:
insignificant extra-solution activity to the judicial exception of data gathering
[using optical character recognition (OCR)] [OCR] [from the OCR]:
merely applying the generic optical character recognition to the abstract idea.
[execute a first processing step based on a first recurrent neural network (RNN), or RNN variant,] [execute a second processing step based on a graph neural network (GNN),or GNN variant, including generating embeddings based on an RNN output from the first RNN] [execute a third processing step using a hybrid convolutional neural network (CNN), the hybrid CNN] [with a CNN][RNN][execute a fourth processing step using a second RNN, or RNN variant, the second RNN modeling the OCR] [ use hybrid backpropagation is used to train the GNN and RNN collaboratively][hybrid CNN] [from the first RNN][to a second RNN][the CNN and first RNN] [CNN and first RNN] [to train the hybrid CNN] [re-training the fully connected neural network][retrain said fully connected neural network]:
merely applying the generic neural networks to the abstract idea.
are recited at a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer components and/or electronic processes. For example, the Applicant’s Specification reads:
[0023] For capturing tokens from the insurance card 102 at OCR module 112, off-the-shelf or in-house optical character recognition systems can be employed.... Several different deep learning architectures can be used, including multimodal architectures which simultaneously process the tokens (character sequences) coming out of the OCR system
[0046] For all the above methods that use neural networks, includes RNNs, CNNs, GNNs, and the like, the training is performed via a gradient descent procedure … can be done through a variety of techniques including but not limited to random normal, Glorot normal, Glorot uniform, and He normal
[0049] While there exist a number of software libraries which assist in the training and deployment of neural networks, one software library that has been found to be advantageous is Tensorflow 2.0 (Google's neural network library), which contains the API needed to construct custom neural networks
[0050] in and out of the deep learning inference pipeline, can be defined as follows….from cloud storage (AWS S3) cloud storage (AWS S3) 314 are loaded into memory using the Tensorflow library
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea and are at a high level of generality. Therefore, Claim 4 is directed to an abstract idea without a practical application. (Step 2A-Prong 2: NO. The additional claimed elements are not integrated into a practical application)
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, the additional elements, do not change the outcome of the analysis, when considered separately and as an ordered combination. The claim further defines the abstract idea and hence is abstract for the reasons presented above. The claim does not include any additional elements that integrate the abstract idea into a practical application or are sufficient to amount to significantly more than the judicial exception when considered both individually and as an ordered combination. Therefore, the claim is directed to an abstract idea. Thus, the claim is not patent eligible. (Step 2B: NO. The claim does not provide significantly more)
Dependent Claims recite additional elements.
This judicial exception is not integrated into a practical application. In particular, the recited additional elements of
Claim 5: (none found: does not include additional elements and merely narrows the abstract idea)
Claim 6: (none found: does not include additional elements and merely narrows the abstract idea)
Claim 7:
“computer”: merely applying computer processing, storage, and networking technology as tools to perform an abstract idea
“training”: merely applying the generic neural networks to the abstract idea
Claim 9:
“CNN”: merely applying the generic neural networks to the abstract idea
are recited at a high-level of generality (i.e., as a generic processor performing a generic computer function) such that it amounts no more than mere instructions to apply the exception using a generic computer components and/or electronic processes. For example, the Applicant’s Specification reads:
[0023] For capturing tokens from the insurance card 102 at OCR module 112, off-the-shelf or in-house optical character recognition systems can be employed.... Several different deep learning architectures can be used, including multimodal architectures which simultaneously process the tokens (character sequences) coming out of the OCR system
[0006] a first recurrent neural network (RNN), or RNN variant,... based on a graph neural network (GNN), or GNN variant
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. For example, the Applicant’s Specification reads:
[0023] For capturing tokens from the insurance card 102 at OCR module 112, off-the-shelf or in-house optical character recognition systems can be employed.... Several different deep learning architectures can be used, including multimodal architectures which simultaneously process the tokens (character sequences) coming out of the OCR system
[0006] a first recurrent neural network (RNN), or RNN variant,... based on a graph neural network (GNN), or GNN variant
[0046] For all the above methods that use neural networks, includes RNNs, CNNs, GNNs, and the like, the training is performed via a gradient descent procedure … can be done through a variety of techniques including but not limited to random normal, Glorot normal, Glorot uniform, and He normal
[0049] While there exist a number of software libraries which assist in the training and deployment of neural networks, one software library that has been found to be advantageous is Tensorflow 2.0 (Google's neural network library), which contains the API needed to construct custom neural networks
[0050] in and out of the deep learning inference pipeline, can be defined as follows….from cloud storage (AWS S3) cloud storage (AWS S3) 314 are loaded into memory using the Tensorflow library
The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, these additional elements, when considered separately and as an ordered combination, do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea and are at a high level of generality. Therefore, the claim is directed to an abstract idea without a practical application. (Step 2A-Prong 2: NO. The additional claimed elements are not integrated into a practical application)
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f). Accordingly, these additional elements, do not change the outcome of the analysis, when considered separately and as an ordered combination. Dependent claims further define the abstract idea that is present in their respective independent claims and hence are abstract for the reasons presented above. The dependent claims do not include any additional elements that integrate the abstract idea into a practical application or are sufficient to amount to significantly more than the judicial exception when considered both individually and as an ordered combination. Therefore, the dependent claims are directed to an abstract idea. Thus, the dependent claims are not patent eligible. (Step 2B: NO. The claims do not provide significantly more)
Claim Rejections - 35 USC § 103
The amended claims facilitated application of additional prior art.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1, 2, 3, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Liao ("TECHNIQUES FOR IMAGE CONTENT EXTRACTION", U.S. Publication Number: 20210366099 A1)in view of Hunter (“GRAPH-MANIPULATION BASED DOMAIN-SPECIFIC EXECUTION ENVIRONMENT”, U.S. Publication Number: 20210073282 A1),in view of Smitherman (“ACCIDENT CLAIMS MANAGEMENT SYSTEM”, U.S. Publication Number: 20140278497 A1), in view of Zadeh (“System and Method for Extremely Efficient Image and Pattern Recognition and Artificial Intelligence Platform”, U.S. Publication Number: 20220121884 A1)
Regarding Claim 1,
Liao teaches,
A system for processing image data generated from a plurality of images of insurance cards to extract discrete entities of information, said system comprising at least one non-transitory computer-readable medium storing software instructions, which, when executed configure one or more processors to receive and process the image data generated from the images of the insurance cards using optical character recognition (OCR) to identify characters on the insurance cards
(Liao [Abstract] extracting contextually structured data from document images, such as by automatically identifying document layout, document data, document metadata
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0041] identify semi-structured data generated by optical character recognition (OCR)
Liao [0277] OCR process performed on the document image may be utilized to identify the set of proximate words
Liao [0247] constructed from various subsets of data
Liao [0007] determine portions of the second textual data
Liao [0143] symbols, characters, terms, numbers
Liao [0302] metadata identified in document images to corresponding data...may identify ‘Name:” ... may identify ‘John Smith’ in the document image as corresponding to the metadata, ‘Name’.
Liao [0269] corresponding locations of the text in each image)
determine a distance between the characters on the insurance cards;
(Liao [0139] blocks of text may be efficiently agglomerated based on spatial proximity
Liao [Claim 5] determine an average line spacing based on a vertical distance between each line in the set of lines
Liao [0041] link words below a threshold distance in the document image)
based, at least in part, on said distance between the characters such that each of the subsets include only the characters within a predetermined distance of one another;
(Liao [0139] blocks of text may be efficiently agglomerated based on spatial proximity
Liao [Claim 5] determine an average line spacing based on a vertical distance between each line in the set of lines
Liao [0041] link words below a threshold distance in the document image together to produce a set of text blocks
Liao [0042] proximity threshold for identification of the set of proximate words for each word token in the set of word tokens comprises a left-direction threshold distance, a right-direction threshold distance, a top-direction threshold distance, and a bottom-direction threshold distance.
Liao [0251] constructed from various subsets of data)
identify relative spatial orientation of said characters of each of said subsets to determine a plurality of tokens, one for each of the subsets of characters and a spatial orientation of said tokens, where the tokens each represent one of a plurality of possible combinations of said characters; emitting the token
(Liao [0277] In many embodiments, word tokens from an OCR process performed on the document image may be utilized to identify the set of proximate words. A set of text blocks may be generated around spatially separate portions of text and adjusted based on proximity and location of data versus metadata. Each of the resulting text blocks may comprise a metadata block with at least one metadata word and a data block with at least one data word corresponding to the metadata block.
Liao [0296] metadata words may be identified utilizing word datasets 1519 corresponding to one or more words and/or combinations of words
Liao [0143] communicated as bits, values, elements, symbols, characters, terms, numbers, or the like.
Liao [0247] constructed from various subsets of data
Liao [0007] determine portions of the second textual data)
on the insurance card;
(Liao [Abstract] extracting contextually structured data from document images, such as by automatically identifying document layout, document data, document metadata
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
such that each of said tokens represents a discrete combination of some or all of the characters
(Liao [0277] In many embodiments, word tokens from an OCR process performed on the document image may be utilized to identify the set of proximate words. A set of text blocks may be generated around spatially separate portions of text and adjusted based on proximity and location of data versus metadata. Each of the resulting text blocks may comprise a metadata block with at least one metadata word and a data block with at least one data word corresponding to the metadata block.
Liao [0296] metadata words may be identified utilizing word datasets 1519 corresponding to one or more words and/or combinations of words
Liao [0143] communicated as bits, values, elements, symbols, characters, terms, numbers, or the like.)
where said plurality of tokens are each modeled as a respective sequence of characters from a set of possible characters including all letters of the alphabet and numbers 0 through 9, and where said plurality of tokens comprise pixel values of the image data;
(Liao [0269] Optical character recognition may be performed on each image...to produce semi-structured data comprising word tokens
Liao [0221] event object may be described as a packet of data...The event object may be created using a variety of formats including ... alphanumeric
Liao [0139] blocks of text may be efficiently agglomerated based ... column-wise pixel intensity)
determine coordinates for each token on the respective one of the insurance cards based on the spatial orientation of the tokens, the tokens and coordinates representing an OCR output;
(Liao [0129] the location comprised in a word token may include a common reference point on a bounding box included in the word token. For example, the bounding box in the word token may include four corners identifying the location of the text in the corresponding image. In such examples, the location used for the linear regression may include the coordinates of one of the four corners of the bounding box
Liao [0341] determined based on corresponding word tokens for each of the document images..... X2 values and Y2 values may correspond to the x-coordinates and y-coordinates, respectively, for locations in the document image
Liao [0041] identify semi-structured data generated by optical character recognition (OCR))
generate a fully connected neural network
(Liao [0317] machine learning algorithms, such as neural networks
Liao [0254] The neural network 1200 is represented as multiple layers of interconnected neurons, such as neuron 1208, that can exchange data between one another.)
based on the image data generated from the insurance cards and the OCR output;
(Liao [Abstract] extracting contextually structured data from document images, such as by automatically identifying document layout, document data, document metadata
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0041] identify semi-structured data generated by optical character recognition (OCR)
Liao [0277] word tokens from an OCR process)
scoring, by way of said fully connected neural network, each node with a … score for the likelihood … corresponds to a … ID for the respective one of the insurance cards;
(Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0333] rankings may be computed based on a composite score derived from the confidence associated with multiple steps/aspects of the process.
Liao [0387] document images using a machine learning (ML) model 3255, such as a neural network
Liao [0390] ML model 3255 may include a neural network, such as one or more of a recurrent neural network (RNN), a convolutional neural network (CNN), a region based CNN, or a Cascade region based CNN
Liao [0059] metadata words and text analytic scoring weights)
for the respective one of the insurance cards; for each of the insurance cards; for said respective one of the insurance cards
(Liao [Abstract] extracting contextually structured data from document images, such as by automatically identifying document layout, document data, document metadata
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
and re-traininq the fully connected network using the stored data
(Liao [0246] During training, input data can be iteratively supplied.... training process may be repeated multiple times
Liao [0254] machine-learning model is the neural network)
wherein said re-training adjusts parameters of the fully connected neural network based on feedback from the stored identified payer names
(Liao [0246] During training, input data can be iteratively supplied to a machine-learning model to enable the machine-learning model to identify patterns related to the input data or to identify relationships between the input data and output data. ...the training process may be repeated multiple times.
Liao [0247] training data is received.
Liao [0248] a machine-learning model is trained using the training data.
Liao [0316] In some embodiments, review manager 1978 may update one or more operational and/or procedural parameters based on corrections, notes, and/or feedback received from users via output manager 1980.
Liao [0322] the metadata database including metadata elements and text analytic scoring weights may be updated. For example, reinforcement trainer 2082 may update metadata elements and text analytic scoring weights to improve future metadata identification and/or correlation.)
and …. to improve subsequent extraction accuracy;
(Liao [0136] Other such embodiments may have increased accuracy and the cost of decreased speed and greater computing demands. Still further embodiments may seek to optimize the cost/performance tradeoff ... This technique may allow more accurate matching with limited loss of compute efficiency.
Liao [0134] This approach differs from pixel-based approaches and can result in improved accuracy, efficiency, and usefulness of data extraction.)
where each of said insurance cards comprise words
(Liao [Abstract] extracting contextually structured data from document images, such as by automatically identifying document layout, document data, document metadata
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0004] identify first text data)
Liao does not teach including a node for each token; member ID; perceived to correspond to a member ID for a respective one of the insurance cards; representing a plausible member ID for the respective one of the insurance cards; node corresponds to the member ID; identifying…a member ID; for said insurance card based on the node with the highest member ID score; scoring each node with a ...name score for the likelihood that said node corresponds to the ...name...and identifying...a ...name... based on the node with the highest ...name score; including the payer name; providing the member ID; of various format; comprising alphanumeric character combinations which are non-correspondent to dictionary words; are provided in different formats; associated with the highest ... score ;ranking labels from a database of predetermined payer labels against the identified payer names; and a highest ranking one of the predetermined payer labels; and storing.... image data, the OCR outputs;
Hunter teaches,
including a node for each token;
(Hunter [0258] a transaction graph edge representing a token exchange
Hunter [0060] a signature value usable by the system to compute a cryptographic hash value.
Hunter [0171] a cryptographic hash of content of the previous node pointed to by an edge connecting those nodes
Examiner considers a token and hash as identical items )
member ID; identifying…a member ID;
(Hunter [0043] a vertex (a term used interchangeably with the term node) of the directed graph may be associated with a norm from a set of norms
Hunter [0101] if the entity “BLMBRG” has an associated entity type of “trusted publisher,” some embodiments may determine that “BLMBRG” is in the set of entities....determining that the entity identifier in the set of entities.
Hunter [0298] the entity identified by ARG2
Examiner notes the Applicant's Specification reads, [0033] "This dataset creation process can be repeated for any entity instead of member ID via the same process" )
node corresponds to the member ID; perceived to correspond to a member ID for a respective one of the insurance cards; representing a plausible member ID for the respective one of the insurance cards;
(Hunter [0043] a vertex (a term used interchangeably with the term node) of the directed graph may be associated with a norm from a set of norms
Hunter [0101] if the entity “BLMBRG” has an associated entity type of “trusted publisher,” some embodiments may determine that “BLMBRG” is in the set of entities....determining that the entity identifier in the set of entities.
Hunter [0298] the entity identified by ARG2
Examiner notes the Applicant's Specification reads, [0033] "This dataset creation process can be repeated for any entity instead of member ID via the same process" )
based on the node
(Hunter [0171] a cryptographic hash of content of the previous node pointed to by an edge connecting those nodes)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the graph-manipulation based domain-specific execution teachings of Hunter “to generate a directed graph and determining a set of triggered vertices based on the directed graph” (Hunter [Abstract]). The modification would have been obvious, because it is merely applying a known technique (i.e. the graph-manipulation based domain-specific execution) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. determine “identifier to characterize a publisher of the event message. … a publisher may be an entity and may include various sources of an event message” Hunter [0056])
Hunter does not teach for said insurance card …with the highest member ID score; scoring each node with a ...name score for the likelihood that said node corresponds to the ...name...and identifying...a ...name... based on the node with the highest ...name score; of various format; comprising alphanumeric character combinations which are non-correspondent to dictionary words; are provided in different formats; associated with the highest ... score ;ranking labels from a database of predetermined payer labels against the identified payer names; and a highest ranking one of the predetermined payer labels; and storing.... image data, the OCR outputs;
Smitherman teaches,
for said insurance card based on the node with the highest member ID score.
(Smitherman [0047] to upload images of patient documentation, such as scans of insurance cards
Smitherman [0098] the payor in the system so that the system can consistently apply rules, consistently track activity
Smitherman [0079] indicates to the user whether a match is a full match or a partial match, or provides another indicator to give the user a likelihood of a match based on matching a name, a date of birth, a medical record, a Social Security number, etc.
Smitherman [0098] once a payor has been identified
Examiner interprets an "indicator to give the user a likelihood of a match" as a "score" ; a "full match" would be the highest score. )
Smitherman explicitly teaches,
perceived to correspond to a member ID for a respective one of the insurance cards; representing a plausible member ID for the respective one of the insurance cards;
(Smitherman [0079] indicates to the user whether a match is a full match or a partial match, or provides another indicator to give the user a likelihood of a match based on matching a name, a date of birth, a medical record, a Social Security number, etc.
Smitherman [0098] once a payor has been identified
Smitherman [0050] other information can be included in the document as well, such as an identification of a payor
Smitherman [0019] a medical record number, a provider name, a patient account number, and dates of service of one of the plurality of candidate patients.
Smitherman [0047] scans of insurance cards)
including the payer name;
(Smitherman [0079] indicates to the user whether a match is a full match or a partial match, or provides another indicator to give the user a likelihood of a match based on matching a name, a date of birth, a medical record, a Social Security number, etc.
Smitherman [0050] other information can be included in the document as well, such as an identification of a payor
Smitherman [0019] a medical record number, a provider name, a patient account number, and dates of service of one of the plurality of candidate patients.)
providing the member ID.
(Smitherman [0053] each patient is assigned a unique number in the system
Smitherman [0079] indicates to the user whether a match is a full match or a partial match, or provides another indicator to give the user a likelihood of a match based on matching a name, a date of birth, a medical record, a Social Security number, etc.
Smitherman [0050] other information can be included in the document as well, such as an identification of a payor
Smitherman [0019] a medical record number, a provider name, a patient account number, and dates of service of one of the plurality of candidate patients.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the accident claims management teachings of Smitherman “in which the received patient is identified as possibly associated with the matching candidate patients.” (Smitherman [0079]). The modification would have been obvious, because it is merely applying a known technique (i.e. accident claims management) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “receiving a document indicating information and a charge, identifying a patient in response to a determination that a portion of the information matches a portion of information of a plurality of candidate patients, and determining an eligibility of the patient for a payor” Smitherman [Abstract])
Smitherman does not teach scoring each node with a ...name score for the likelihood that said node corresponds to the ...name...and identifying...a ...name... based on the node with the highest ...name score; of various format; comprising alphanumeric character combinations which are non-correspondent to dictionary words; are provided in different formats; associated with the highest ... score ;ranking labels from a database of predetermined payer labels against the identified payer names; and a highest ranking one of the predetermined payer labels; and storing.... image data, the OCR outputs;
Zedah teaches,
scoring each node with a ...name score for the likelihood that said node corresponds to the ...name...and identifying...a ...name... based on the node with the highest ...name score; associated with the highest ... score ;
(Zadeh [0759] similar fuzzy description of music is used to determine/search/find the candidates...with best match(es) and/or rankings
Zadeh [1505] the system ranks the results, and it marks the result that has the highest score in the context of the phrase or sentence, for possible candidate for the original (correct) word
Zadeh [2125] for example, using matching or recognition confidence or score
Zadeh [2756] it should be more relevant and focused, and it should come up to the top of the list, with a higher score or ranking)
ranking labels from a database of predetermined payer labels against the identified payer names;
(Zadeh [0759] (or metadata) with best match(es) and/or rankings
Zadeh [1438] names and titles are stored and indexed or ranked
Zadeh [1502] “Clinton” (extracted from the sentence and distinguished as a possible given name or family name) is already stored in a database
Zadeh [2002] having at least one of each classification stored for future referral.
Zadeh [2510] customer profile)
and a highest ranking one of the predetermined payer labels;
(Zadeh [1505] system ranks the results, and it marks the result that has the highest score... FIG. 88 shows an example of such system.
Zadeh [2756] the system sorts inversely, .... indicates more specific object(s), or topic, and so, it should be more relevant and focused, and it should come up to the top of the list, with a higher score or ranking.)
and storing.... image data, the OCR outputs;
(Zadeh [0759] similar fuzzy description of music is used to determine/search/find the candidates...with best match(es) and/or rankings
Zadeh [1505] the system ranks the results, and it marks the result that has the highest score in the context of the phrase or sentence, for possible candidate for the original (correct) word
Zadeh [2125] for example, using matching or recognition confidence or score
Zadeh [2756] it should be more relevant and focused, and it should come up to the top of the list, with a higher score or ranking)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the highest score rankings of Zadeh “system ranks the results, and it marks the result that has the highest score.” (Zadeh [1505]). The modification would have been obvious, because it is merely applying a known technique (i.e. highest score rankings ) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “(or metadata) with best match(es) and/or rankings” Zadeh [0759])
Regarding Claim 2,
Liao, Hunter, Smitherman, and Zadeh teach the entity information extraction from insurance card images of Claim 1 as described earlier.
Liao teaches,
wherein generating the fully connected neural network includes modeling the OCR output for each of the insurance cards by using vector representations of each token.
(Liao [0317] machine learning algorithms, such as neural networks
Liao [0254] The neural network 1200 is represented as multiple layers of interconnected neurons, such as neuron 1208, that can exchange data between one another.
Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0271] a binary vector may be generated for each page based on the list of common words. For example, each dimension in the binary vector may correspond to a word in the list
Liao [0276] binary vectors (e.g., representing presence of common words in a template) may be utilized)
Regarding Claim 3,
Liao, Hunter, Smitherman, and Zadeh teach the entity information extraction from insurance card images of Claim 1 as described earlier.
Liao teaches,
generating the fully connected neural network includes generating a graph based on the OCR output
(Liao [0216] An event stream processing engine (ESPE) may continuously apply the queries to the data
Liao [0220] modeling involves defining directed graphs of windows for event stream manipulation and transformation.... A directed graph, for example, is a set of nodes connected by edges, where the edges have a direction associated with them.
Liao [0041] identify semi-structured data generated by optical character recognition (OCR)
Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0254] The neural network 1200 is represented as multiple layers of interconnected neurons, such as neuron 1208, that can exchange data between one another.)
for each of the insurance cards,
(Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
with each token for said respective one of the insurance cards taken as a node of the graph and edges being declared when … (distance is below a given threshold)
(Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0216] An event stream processing engine (ESPE) may continuously apply the queries to the data
Liao [0220] modeling involves defining directed graphs of windows for event stream manipulation and transformation.... A directed graph, for example, is a set of nodes connected by edges, where the edges have a direction associated with them.
Liao [0007] determine an average line spacing based on a vertical distance between each line in the set of lines
Liao [0022] compute a hamming distance between each pair of image hashes
Liao [0041] link words below a threshold distance in the document image together to produce a set of text blocks)
on the card
(Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
Liao does not teach Euclidean distance between token centroids is below a given threshold and rejecting edges when the Euclidean distance between the token centroids is above the threshold, where the threshold is selected to reflect physical spacing...; a node feature matrix is constructed based on the graph; and scoring each node is based, at least in part, on the node feature matrix.
Zedah teaches,
Euclidean distance between token centroids is below a given threshold and rejecting edges when the Euclidean distance between the token centroids is above the threshold, where the threshold is selected to reflect physical spacing...;
(Zadeh [2351] we use one of the distance metrics, e.g. Euclidean distance between 2 points.... and cluster centroid or center of mass or average value
Zadeh [2653] statistical threshold is used to determine whether the mismatch is attributed to jitter or a consistent bias (e.g., out of synch issue).
Zadeh [0325] based on some thresholds or conditions
Zadeh [3304] failure (not finding feature)
Zadeh [1477] The recognition of...text, including OCR (optical character recognition), is generally done by dissecting the image or video into pieces and components... to find features or objects, and from the parameters associated ..., e.g. geometrical lengths or ratios or angles, the system finds or guesses the identity of those features or objects
Zadeh [2958] to carry the spatial information (e.g., ... an outline of the object...) and the attributes such as the coordinates and types of the spatial indicator (outline, rectangle, ellipse/oval, region/blob/silhouette of the object))
a node feature matrix is constructed based on the graph; and scoring each node is based, at least in part, on the node feature matrix.
(Zadeh [1702] from training data, apply discriminant filter to rules, and create a combined fuzzy associative memory (FAM), which is a matrix (based on the inputs).
Zadeh [1818] training dataset to let the system learn features...with various parameters,...the system learn the invariant features... orthogonal matrixes
Zadeh [1757] the decomposition of B.sub.x may be expressed as series of tuples in the form (V.sub.j,TS.sub.k,weig,ht.sub.j,k) or simply as a matrix with weight.sub.j,k as its elements. Given the correspondence between C.sub.j and V.sub.j, the granular test score sets TS.sub.k's are also associated
Zadeh [2633] affine transformations for geometrical matching ... gets matching scores, which indicate the result of comparisons
Zadeh [1864] a set of feature nodes/units/neurons are added, e.g., to the top hidden layer of RBMs, for training to detect features
Zadeh [2880] the hierarchical algorithm, or tree decision making, for which each node is a recognizer of a feature)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the highest score rankings of Zadeh “system ranks the results, and it marks the result that has the highest score.” (Zadeh [1505]). The modification would have been obvious, because it is merely applying a known technique (i.e. highest score rankings ) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “(or metadata) with best match(es) and/or rankings” Zadeh [0759])
Regarding Claim 19,
Liao, Hunter, Smitherman, and Zadeh teach the entity information extraction from insurance card images of Claim 1 as described earlier.
Liao teaches,
insurance cards
(Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
Liao does not teach generate a database registering one member …, each registration including the identified member … and the highest ranking one of the predetermined payer labels for each registered member.
Hunter teaches
member ID;
(Hunter [0043] a vertex (a term used interchangeably with the term node) of the directed graph may be associated with a norm from a set of norms
Hunter [0101] if the entity “BLMBRG” has an associated entity type of “trusted publisher,” some embodiments may determine that “BLMBRG” is in the set of entities....determining that the entity identifier in the set of entities.
Hunter [0298] the entity identified by ARG2
Examiner notes the Applicant's Specification reads, [0033] "This dataset creation process can be repeated for any entity instead of member ID via the same process" )
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the graph-manipulation based domain-specific execution teachings of Hunter “to generate a directed graph and determining a set of triggered vertices based on the directed graph” (Hunter [Abstract]). The modification would have been obvious, because it is merely applying a known technique (i.e. the graph-manipulation based domain-specific execution) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. determine “identifier to characterize a publisher of the event message. … a publisher may be an entity and may include various sources of an event message” Hunter [0056])
Hunter does not teach generate a database registering one member …, each registration including the identified member … and the highest ranking one of the predetermined payer labels for each registered member.
Zedah teaches,
generate a database registering one member …, each registration including the identified member … and the highest ranking one of the predetermined payer labels for each registered member.
(Zadeh [0759] (or metadata) with best match(es) and/or rankings
Zadeh [1438] names and titles are stored and indexed or ranked
Zadeh [1505] the system ranks the results, and it marks the result that has the highest score in the context of the phrase or sentence, for possible candidate for the original (correct) word
Zadeh [2125] for example, using matching or recognition confidence or score
Zadeh [2756] it should be more relevant and focused, and it should come up to the top of the list, with a higher score or ranking
Zadeh [1502] “Clinton” (extracted from the sentence and distinguished as a possible given name or family name) is already stored in a database
Zadeh [2002] having at least one of each classification stored for future referral.
Zadeh [2510] customer profile)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the highest score rankings of Zadeh “system ranks the results, and it marks the result that has the highest score.” (Zadeh [1505]). The modification would have been obvious, because it is merely applying a known technique (i.e. highest score rankings ) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “(or metadata) with best match(es) and/or rankings” Zadeh [0759])
Claims 4-7, 9, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Liao, Hunter, Smitherman, and Zadeh in view of Collins (“TECHNIQUES FOR INFERRING INFORMATION”, U.S. Publication Number: 20230123811 A1)
Regarding Claim 4,
Liao teaches,
A system for processing a plurality of insurance cards …each of the insurance cards including a payer name … payer names … locations across the insurance cards, said system comprising:
(Liao [0269] corresponding locations of the text in each image
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0302] metadata identified in document images to corresponding data...may identify ‘Name:” ... may identify ‘John Smith’ in the document image as corresponding to the metadata, ‘Name’. Accordingly, contextually structured data extracted from the document image may include a key-value pair of ‘Name’-‘John Smith’.))
a camera configured to capture a plurality of images, the images including at least one image corresponding to each one of the insurance cards, each of said insurance cards each comprising a set of words; at least one computer configured, for each insurance card,
(Liao [0120] an OCR process
Liao [0423] cameras or camera arrays
Liao [Abstract] extracting contextually structured data from document images
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0004] identify first text data
Liao [0010] hereby include a computer-implemented method
Liao [0247] constructed from various subsets of data
Liao [0007] determine portions of the second textual data
Liao [0143] symbols, characters, terms, numbers
Liao [0143] communicated as bits, values, elements, symbols, characters, terms, numbers, or the like.)
receive and process image data generated from the at least one image of the insurance card using optical character recognition (OCR) to identify characters on the insurance cards;
(Liao [0041] identify semi-structured data generated by optical character recognition (OCR)
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
based, at least in part, on said distance between the characters such that each of the subsets include only the characters within a predetermined distance of one another;
(Liao [0139] blocks of text may be efficiently agglomerated based on spatial proximity
Liao [Claim 5] determine an average line spacing based on a vertical distance between each line in the set of lines
Liao [0041] link words below a threshold distance in the document image together to produce a set of text blocks
Liao [0042] proximity threshold for identification of the set of proximate words for each word token in the set of word tokens comprises a left-direction threshold distance, a right-direction threshold distance, a top-direction threshold distance, and a bottom-direction threshold distance.
Liao [0251] constructed from various subsets of data)
identify relative spatial orientation of said characters to determine a plurality of tokens, one for each of said subsets of characters and a spatial orientation of said tokens, where the tokens each represent one of a plurality of possible combinations of said characters on the insurance card;
(Liao [0277] In many embodiments, word tokens from an OCR process performed on the document image may be utilized to identify the set of proximate words. A set of text blocks may be generated around spatially separate portions of text and adjusted based on proximity and location of data versus metadata. Each of the resulting text blocks may comprise a metadata block with at least one metadata word and a data block with at least one data word corresponding to the metadata block.
Liao [0296] metadata words may be identified utilizing word datasets 1519 corresponding to one or more words and/or combinations of words
Liao [0143] communicated as bits, values, elements, symbols, characters, terms, numbers, or the like.
Liao [0247] constructed from various subsets of data
Liao [0007] determine portions of the second textual data)
where said plurality of tokens are modeled as a respective sequence of characters from a set of possible characters including all letters of the alphabet and numbers 0 through 9, and where said plurality of tokens comprise pixel values of the image data;
(Liao [0269] Optical character recognition may be performed on each image...to produce semi-structured data comprising word tokens
Liao [0221] event object may be described as a packet of data...The event object may be created using a variety of formats including ... alphanumeric
Liao [0139] blocks of text may be efficiently agglomerated based ... column-wise pixel intensity)
such that each of said tokens represents a discrete combination of some or all of the subset of characters
(Liao [0277] In many embodiments, word tokens from an OCR process performed on the document image may be utilized to identify the set of proximate words. A set of text blocks may be generated around spatially separate portions of text and adjusted based on proximity and location of data versus metadata. Each of the resulting text blocks may comprise a metadata block with at least one metadata word and a data block with at least one data word corresponding to the metadata block.
Liao [0296] metadata words may be identified utilizing word datasets 1519 corresponding to one or more words and/or combinations of words
Liao [0143] communicated as bits, values, elements, symbols, characters, terms, numbers, or the like.)
determine coordinates for each token on the insurance card based on the spatial orientation of the tokens;
(Liao [0129] the location comprised in a word token may include a common reference point on a bounding box included in the word token. For example, the bounding box in the word token may include four corners identifying the location of the text in the corresponding image. In such examples, the location used for the linear regression may include the coordinates of one of the four corners of the bounding box
Liao [0341] determined based on corresponding word tokens for each of the document images..... X2 values and Y2 values may correspond to the x-coordinates and y-coordinates, respectively, for locations in the document image
Liao [0041] identify semi-structured data generated by optical character recognition (OCR))
execute a first processing step based on a first recurrent neural network (RNN), or RNN variant, to model the OCR output for each of the plurality of insurance cards, using vector representations of each token
(Liao [0390] For example, ML model 3255 may include a neural network, such as one or more of a recurrent neural network (RNN)
Liao [0041] identify semi-structured data generated by optical character recognition (OCR)
Liao [0271] a binary vector may be generated for each page based on the list of common words. For example, each dimension in the binary vector may correspond to a word in the list
Liao [0276] binary vectors (e.g., representing presence of common words in a template) may be utilized
Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image)
and the OCR output to generate a graph
(Liao [0041] identify semi-structured data generated by optical character recognition (OCR)
Liao [0220] modeling involves defining directed graphs of windows for event stream manipulation and transformation.... A directed graph, for example, is a set of nodes connected by edges, where the edges have a direction associated with them.)
execute a third processing step using a hybrid convolutional neural network (CNN), the hybrid CNN processing the image of each of the plurality of insurance cards with a CNN to generate an image representation of each insurance card
(Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0009] one or more of the first and second ML models comprise a recurrent neural network, a convolutional neural network (CNN), a region based CNN, or a Cascade region based CNN.
Liao [0220] modeling involves defining directed graphs of windows for event stream manipulation and transformation.... A directed graph, for example, is a set of nodes connected by edges, where the edges have a direction associated with them.)
and combining each image representation with a hidden output from the first RNN;
(Liao [0297] in document images to identify metadata
Liao [0162] high value analytics can be applied to identify hidden relationships
Liao [0258] The transformed output can be supplied to a subsequent layer, such as the hidden layer 1204, of the neural network 1200. The subsequent layer of the neural network 1200 can receive the transformed output, multiply the transformed output by a matrix of numeric weights and a nonlinearity, and provide the result to yet another layer of the neural network)
for each of the plurality of insurance cards,
(Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
an image embedding of the card
(Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
execute a fourth processing step using the RNN variant, the RNN variant modeling the OCR output using the fixed length vector from the image of the insurance card;
(Liao [0050] data generated by optical character recognition (OCR)
Liao [0390] such as one or more of a recurrent neural network (RNN)
Liao [0009] one or more of the first and second ML models comprise a recurrent neural network, a convolutional neural network (CNN), a region based CNN, or a Cascade region based CNN.
Liao [0199] fixed time intervals, between known fixed stages
Liao [0166] characteristics that may be sensed include... distance, speed, vibrations, acceleration, ... among others.
Liao [0270] the hamming distance between each of the image hashes)
assigning a score to the tokens based on a likelihood that the token corresponds
(Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0333] rankings may be computed based on a composite score derived from the confidence associated with multiple steps/aspects of the process.)
and the payer name, respectively; comprising the payor name … for re-training the fully connected neural network; retrain said fully connected neural network using said stored image data, output from the OCR, and predictive output comprising the payor name and the member ID
(Liao [0041] data generated by optical character recognition (OCR)
Liao [0246] During training, input data can be iteratively supplied.... training process may be repeated multiple times
Liao [0254] machine-learning model is the neural network
Liao [0302] metadata identified in document images to corresponding data...may identify ‘Name:” ... may identify ‘John Smith’ in the document image as corresponding to the metadata, ‘Name’. Accordingly, contextually structured data extracted from the document image may include a key-value pair of ‘Name’-‘John Smith’.) )
and transmit discrete entity of information …from (each of the plurality of insurance cards to a remote electronic device for display or further processing)…each associated with one or more of the insurance cards
(Liao [0152] Data transmission network 100 also includes one or more network devices
[0160] or remote server
Liao [0041] identify semi-structured data
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0423] one or more displays to present information)
Liao does not teach to obtain a logit for each token; including generating embeddings based on an RNN output from the first RNN, the embeddings being vector representations of the tokens, and using the embeddings; with each token as a node, to construct a node feature matrix; execute a second processing step based on a graph neural network (GNN), or GNN variant; extract at least one discrete entity of information … based on the processing steps; perceived to correspond to a member ID; including at least a member ID; corresponding to a plausible member ID for the insurance card; the member ID; a alphanumeric combination of spaced characters providing a member identifier (ID); of various format; comprising alphanumeric characters which are non-correspondent to dictionary words... are provided in different formats; computer comprising a database of predetermined payer labels and software instructions, which when executed, configure the at least one computer; and use hybrid backpropagation is used to train the GNN and RNN collaboratively; wherein the hybrid CNN concatenates... with a hidden state from the first RNN, the concatenation forming a fixed length vector input to a second RNN, and wherein the CNN and first RNN are joined and the parameters of the CNN and first RNN are updated simultaneously to train the hybrid CNN; from and the highest ranking one of the predetermined payer labels ; generate a database of information for a plurality of members... the database registering one member....and including the highest ranking one of the predetermined payer labels name and the discrete entity of information associated with the member ID for each registered member
Hunter teaches,
member ID,
(Hunter [0101] if the entity “BLMBRG” has an associated entity type of “trusted publisher,” some embodiments may determine that “BLMBRG” is in the set of entities....determining that the entity identifier in the set of entities.
Hunter [0298] the entity identified by ARG2)
execute a second processing step based on a graph neural network (GNN), or GNN variant,
(Hunter [0230] the deep neural network may be a deep convolutional neural network, deep graph neural network)
including generating embeddings based on an RNN output from the first RNN, the embeddings being vector representations of the tokens, and using the embeddings
(Hunter [0360] determine outcome determination parameters based on a directed graph may include embeddings based on the directed graph
Hunter [0230] the deep neural network may be a deep convolutional neural network, deep graph neural network, deep recurrent neural network, some combination thereof, or the like.
Hunter [0095] AI systems include systems that may determine a set of outputs from a set of inputs ...or other interpretable AI systems
Hunter [0220] data structure instances (e.g., an array, a vector, a list, or the like)
Hunter [0258] a transaction graph edge representing a token exchange)
with each token as a node, to construct a node feature matrix;
(Hunter [0362] a first entity may include or be based on a feature matrix representing whether other entities
Hunter [0258] a transaction graph edge representing a token exchange
Hunter [0060] a signature value usable by the system to compute a cryptographic hash value.
Hunter [0171] a cryptographic hash of content of the previous node pointed to by an edge connecting those nodes
Examiner considers a token and hash as identical items )
extract at least one discrete entity of information … based on the processing steps by
(Hunter [0101] if the entity “BLMBRG” has an associated entity type of “trusted publisher,” some embodiments may determine that “BLMBRG” is in the set of entities....determining that the entity identifier in the set of entities.
Hunter [0298] the entity identified by ARG2)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the graph-manipulation based domain-specific execution teachings of Hunter “to generate a directed graph and determining a set of triggered vertices based on the directed graph” (Hunter [Abstract]). The modification would have been obvious, because it is merely applying a known technique (i.e. the graph-manipulation based domain-specific execution) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. determine “identifier to characterize a publisher of the event message. … a publisher may be an entity and may include various sources of an event message” Hunter [0056])
Hunter does not teach to obtain a logit for each token; including at least a member ID; a alphanumeric combination of evenly spaced characters providing a member ID; of various format; comprising alphanumeric characters which are non-correspondent to dictionary words... are provided in different formats; computer comprising a database of predetermined payer labels and software instructions, which when executed, configure the at least one computer; and use hybrid backpropagation is used to train the GNN and RNN collaboratively; wherein the hybrid CNN concatenates... with a hidden state from the first RNN, the concatenation forming a fixed length vector input to a second RNN, and wherein the CNN and first RNN are joined and the parameters of the CNN and first RNN are updated simultaneously to train the hybrid CNN; and comparing the discrete entity of information associated with the payer name against the database of predetermined payer labels and ranking a likelihood of each of the predetermined payer labels; store said image data, output from the OCR, and predictive output from and the highest ranking one of the predetermined payer labels ; generate a database of information for a plurality of members... the database registering one member....and including the highest ranking one of the predetermined payer labels name and the discrete entity of information … for each registered member
Smitherman teaches,
an alphanumeric combination of evenly spaced characters providing a member ID;
(Smitherman [0053] each patient is assigned a unique number in the system
Smitherman [0079] indicates to the user whether a match is a full match or a partial match, or provides another indicator to give the user a likelihood of a match based on matching a name, a date of birth, a medical record, a Social Security number, etc.
Smitherman [0050] other information can be included in the document as well, such as an identification of a payor
Smitherman [0019] a medical record number, a provider name, a patient account number, and dates of service of one of the plurality of candidate patients.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the accident claims management teachings of Smitherman “in which the received patient is identified as possibly associated with the matching candidate patients.” (Smitherman [0079]). The modification would have been obvious, because it is merely applying a known technique (i.e. accident claims management) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “receiving a document indicating information and a charge, identifying a patient in response to a determination that a portion of the information matches a portion of information of a plurality of candidate patients, and determining an eligibility of the patient for a payor” Smitherman [Abstract])
Smitherman does not teach to obtain a logit for each token; of various format; comprising alphanumeric characters which are non-correspondent to dictionary words... are provided in different formats; computer comprising a database of predetermined payer labels and software instructions, which when executed, configure the at least one computer; and use hybrid backpropagation is used to train the GNN and RNN collaboratively; wherein the hybrid CNN concatenates... with a hidden state from the first RNN, the concatenation forming a fixed length vector input to a second RNN, and wherein the CNN and first RNN are joined and the parameters of the CNN and first RNN are updated simultaneously to train the hybrid CNN; and comparing the discrete entity of information associated with the payer name against the database of predetermined payer labels and ranking a likelihood of each of the predetermined payer labels; store said image data, output from the OCR, and predictive output from and the highest ranking one of the predetermined payer labels ; generate a database of information for a plurality of members... the database registering one member....and including the highest ranking one of the predetermined payer labels name and the discrete entity of information … for each registered member
Zedah teaches,
computer comprising a database of predetermined payer labels and software instructions, which when executed, configure the at least one computer;
(Zadeh [1438] names and titles are stored and indexed or ranked
Zadeh [1502] “Clinton” (extracted from the sentence and distinguished as a possible given name or family name) is already stored in a database
Zadeh [1438] names and titles are stored and indexed or ranked
Zadeh [1502] “Clinton” (extracted from the sentence and distinguished as a possible given name or family name) is already stored in a database)
and use hybrid backpropagation is used to train the GNN and RNN collaboratively;
(Zadeh [0573] for cognition layer for complex hybrid data for our/ZAC Explainable-AI system
Zadeh [2260] we use backpropagation method
Zadeh [1486] a multi-layer feed-forward neural network
Zadeh [1795] a type of stochastic recurrent neural network)
and comparing the discrete entity of information associated with the payer name against the database of predetermined payer labels and ranking a likelihood of each of the predetermined payer labels;
(Zadeh [0759] (or metadata) with best match(es) and/or rankings
Zadeh [1438] names and titles are stored and indexed or ranked
Zadeh [1502] “Clinton” (extracted from the sentence and distinguished as a possible given name or family name) is already stored in a database
Zadeh [2002] having at least one of each classification stored for future referral.
Zadeh [2510] customer profile)
store said image data, output from the OCR, and predictive output ….from and the highest ranking one of the predetermined payer labels ;
(Zadeh [1436] an image or picture is stored at different resolutions (with different sizes) at different repositories.
Zadeh [1477] The recognition of ... text, including OCR (optical character recognition), is generally done by dissecting the image or video into pieces and components... to find features or objects, and from the parameters associated with those features and objects
Zadeh [1946] which is connected to output module
Zadeh [0759] similar fuzzy description of music is used to determine/search/find the candidates...with best match(es) and/or rankings
Zadeh [1505] the system ranks the results, and it marks the result that has the highest score in the context of the phrase or sentence, for possible candidate for the original (correct) word
Zadeh [2125] for example, using matching or recognition confidence or score
Zadeh [2756] it should be more relevant and focused, and it should come up to the top of the list, with a higher score or ranking,)
to obtain a logit for each token
(Zadeh [2443] In one embodiment, we use text normalization or sentence tokenization
Zadeh [1803] the visible units have continuous value state (e.g., logistic units)...and signals from its corresponding visible units, V.sup.(3) (e.g., logistic units).)
generate a database of information for a plurality of members... the database registering one member....and including the highest ranking one of the predetermined payer labels name and the discrete entity of information … for each registered member
(Zadeh [0759] similar fuzzy description of music is used to determine/search/find the candidates...with best match(es) and/or rankings
Zadeh [1505] the system ranks the results, and it marks the result that has the highest score in the context of the phrase or sentence, for possible candidate for the original (correct) word
Zadeh [0759] (or metadata) with best match(es) and/or rankings
Zadeh [1438] names and titles are stored and indexed or ranked
Zadeh [1502] “Clinton” (extracted from the sentence and distinguished as a possible given name or family name) is already stored in a database
Zadeh [2002] having at least one of each classification stored for future referral.
Zadeh [2510] customer profile)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the highest score rankings of Zadeh “system ranks the results, and it marks the result that has the highest score.” (Zadeh [1505]). The modification would have been obvious, because it is merely applying a known technique (i.e. highest score rankings ) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “(or metadata) with best match(es) and/or rankings” Zadeh [0759])
Zedah does not teach wherein the hybrid CNN concatenates... with a hidden state from the first RNN, the concatenation forming a fixed length vector input to a second RNN, and wherein the CNN and first RNN are joined and the parameters of the CNN and first RNN are updated simultaneously to train the hybrid CNN;
Collins teaches,
wherein the hybrid CNN concatenates
(Collins [0593] may include one or more machine learning models (e.g., convolutional neural networks (CNNs))
Collins [0125] a current state and outputs a next state, x is an input tensor, h_in is an initial hidden state, h_out is a final hidden state
Collins [0440] can additionally include one or more fixed function or special function units to perform specific functions
Collins [0063] include specific values (e.g., [256] for a one-dimensional vector,)
with a hidden state from the first RNN, the concatenation forming a fixed length vector
(Collins [0593] may include one or more machine learning models (e.g., convolutional neural networks (CNNs))
Collins [0125] a current state and outputs a next state, x is an input tensor, h_in is an initial hidden state, h_out is a final hidden state
Collins [0440] can additionally include one or more fixed function or special function units to perform specific functions
Collins [0063] include specific values (e.g., [256] for a one-dimensional vector,)
input to a second RNN,
(Collins [0593] may include one or more machine learning models (e.g., convolutional neural networks (CNNs))
Collins [0062] convolutional neural network (CNN), and/or a recurrent neural network (RNN)
Collins [0267] allow for multiple neural networks to be performed simultaneously and/or sequentially, and for results to be combined together
Collins [0272] used to train and/or update neural networks based at least in part on input
Collins [0308] neural networks 1492, update
Collins [0521] fixed function matrix multiplication logic, for implementations including optimizations for machine learning training or inferencing.)
and wherein the CNN and first RNN are joined and the parameters of the CNN and first RNN
(Collins [Abstract] uses one or more neural networks to infer information
Collins [0062] convolutional neural network (CNN), and/or a recurrent neural network (RNN)
Collins [0601] image processing request....container, dataset, machine learning model, etc., select a desired combination of elements for inclusion in data processing pipeline
Collins [0135] one or more inputs to a neural network graph,...determine one or more dimensions based, at least in part, on a specific value of a dimension of a tensor used as an output in neural network graph. ... based, at least in part, on one or more rules associated with one or more of a concatenation operation, a matrix multiplication operation, and a convolution operation)
are updated simultaneously to train the hybrid CNN;
(Collins [0267] allow for multiple neural networks to be performed simultaneously and/or sequentially, and for results to be combined together
Collins [0272] used to train and/or update neural networks based at least in part on input
Collins [0308] neural networks 1492, update
Collins [0521] fixed function matrix multiplication logic, for implementations including optimizations for machine learning training or inferencing.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate update multiple neural networks of Collins for “allow for multiple neural networks to be performed simultaneously and/or sequentially, and for results to be combined together” (Collins [0267]). The modification would have been obvious, because it is merely applying a known technique (i.e. update multiple neural networks) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “used to train and/or update neural networks based at least in part on input” Collins [0272])
Regarding Claim 5,
Liao, Hunter, Smitherman, Zadeh, and Collins teach the entity information extraction from insurance card images of Claim 4 as described earlier.
Liao teaches,
when executing the second processing step, edges define connections between nodes on the graph when a … distance between token centroids does not exceed predetermined threshold.
(Liao [0220] modeling involves defining directed graphs of windows for event stream manipulation and transformation.... A directed graph, for example, is a set of nodes connected by edges, where the edges have a direction associated with them.
Liao [0007] determine an average line spacing based on a vertical distance between each line in the set of lines
Liao [0022] compute a hamming distance between each pair of image hashes...exceed a threshold
Liao [0041] threshold distance in the document image together to produce a set of text blocks
Examiner understand the inverse remains true.)
Liao does not teach rejecting edges when the Euclidean distance between the token centroids is above the threshold, where the threshold is selected to reflect physical spacing...;
Zedah teaches,
rejecting edges when the Euclidean distance between the token centroids exceeds the predetermined threshold, where the threshold is selected to reflect physical spacing...;
(Zadeh [2351] we use one of the distance metrics, e.g. Euclidean distance between 2 points.... and cluster centroid or center of mass or average value
Zadeh [2653] statistical threshold is used to determine whether the mismatch is attributed to jitter or a consistent bias (e.g., out of synch issue).
Zadeh [0325] based on some thresholds or conditions
Zadeh [3304] failure (not finding feature)
Zadeh [1477] The recognition of...text, including OCR (optical character recognition), is generally done by dissecting the image or video into pieces and components... to find features or objects, and from the parameters associated ..., e.g. geometrical lengths or ratios or angles, the system finds or guesses the identity of those features or objects
Zadeh [2958] to carry the spatial information (e.g., ... an outline of the object...) and the attributes such as the coordinates and types of the spatial indicator (outline, rectangle, ellipse/oval, region/blob/silhouette of the object)
Examiner understand the inverse remains true)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the highest score rankings of Zadeh “system ranks the results, and it marks the result that has the highest score.” (Zadeh [1505]). The modification would have been obvious, because it is merely applying a known technique (i.e. highest score rankings ) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “(or metadata) with best match(es) and/or rankings” Zadeh [0759])
Regarding Claim 6,
Liao, Hunter, Smitherman, Zadeh, and Collins teach the entity information extraction from insurance card images of Claim 5 as described earlier.
Liao teaches,
to with one of a plurality of expected characteristics comprising …the payer name; and during the step of extracting the at least one discrete entity of information, the score for each entity is assigned to each token
(Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0269] produce semi-structured data comprising word tokens that include text identified in each image and the corresponding locations of the text in each image
Liao [0333] rankings may be computed based on a composite score derived from the confidence associated with multiple steps/aspects of the process.
Liao [0302] metadata identified in document images to corresponding data...may identify ‘Name:” ... may identify ‘John Smith’ in the document image as corresponding to the metadata, ‘Name’. Accordingly, contextually structured data extracted from the document image may include a key-value pair of ‘Name’-‘John Smith’.))
Liao does not teach member ID; each processing step generates at least one logit for each token correlating said token; based on the logits correlating said token to one of the expected characteristics.
Zedah teaches,
each processing step generates at least one logit for each token correlating said token; based on the logits correlating said token to one of the expected characteristics.
(Zadeh [2443] In one embodiment, we use text normalization or sentence tokenization
Zadeh [1803] the visible units have continuous value state (e.g., logistic units)...and signals from its corresponding visible units, V.sup.(3) (e.g., logistic units).)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the highest score rankings of Zadeh “system ranks the results, and it marks the result that has the highest score.” (Zadeh [1505]). The modification would have been obvious, because it is merely applying a known technique (i.e. highest score rankings ) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “(or metadata) with best match(es) and/or rankings” Zadeh [0759])
Zedah does not teach member ID;
Hunter teaches
member ID;;
(Hunter [0043] a vertex (a term used interchangeably with the term node) of the directed graph may be associated with a norm from a set of norms
Hunter [0101] if the entity “BLMBRG” has an associated entity type of “trusted publisher,” some embodiments may determine that “BLMBRG” is in the set of entities....determining that the entity identifier in the set of entities.
Hunter [0298] the entity identified by ARG2
Examiner notes the Applicant's Specification reads, [0033] "This dataset creation process can be repeated for any entity instead of member ID via the same process" )
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the graph-manipulation based domain-specific execution teachings of Hunter “to generate a directed graph and determining a set of triggered vertices based on the directed graph” (Hunter [Abstract]). The modification would have been obvious, because it is merely applying a known technique (i.e. the graph-manipulation based domain-specific execution) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. determine “identifier to characterize a publisher of the event message. … a publisher may be an entity and may include various sources of an event message” Hunter [0056])
Regarding Claim 7,
Liao, Hunter, Smitherman, Zadeh, and Collins teach the entity information extraction from insurance card images of Claim 6 as described earlier.
Liao teaches,
wherein the at least one computer is further configured to train the system during the processing steps by executing the processing steps on insurance cards
(Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0246] Machine-learning models can be constructed through an at least partially automated (e.g., with little or no human involvement) process called training. During training, input data can be iteratively supplied)
comprising: a first group of insurance cards representing a validation set; and a second group of insurance cards representing a training set.
(Liao [0246] Input data can be split into one or more training sets and one or more validation sets)
Regarding Claim 9,
Liao, Hunter, Smitherman, Zadeh, and Collins teach the entity information extraction from insurance card images of Claim 7 as described earlier.
Liao teaches,
payer name
(Liao [0302] metadata identified in document images to corresponding data...may identify ‘Name:” ... may identify ‘John Smith’ in the document image as corresponding to the metadata, ‘Name’. Accordingly, contextually structured data extracted from the document image may include a key-value pair of ‘Name’-‘John Smith’.))
with the CNN during execution of the third processing step.
(Liao [0390] may include a neural network, such as one or more of a recurrent neural network (RNN), a convolutional neural network (CNN), a region based CNN, or a Cascade region based CNN. Generally, object detector 3236 may include one or more ML models
Liao [0333] rankings may be computed based on a composite score derived from the confidence associated with multiple steps/aspects of the process.)
Liao does not teach wherein the system determines the payer
Smitherman teaches,
wherein the system determines the payer
(Smitherman [0043] to determine primary payors to avoid paying a claim out of order. A subrogation entity can use the system to confirm whether there is a liability claim and/or a no-fault claim pending so that a payor does not overpay its portion of responsibility.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the accident claims management teachings of Smitherman “in which the received patient is identified as possibly associated with the matching candidate patients.” (Smitherman [0079]). The modification would have been obvious, because it is merely applying a known technique (i.e. accident claims management) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “receiving a document indicating information and a charge, identifying a patient in response to a determination that a portion of the information matches a portion of information of a plurality of candidate patients, and determining an eligibility of the patient for a payor” Smitherman [Abstract])
Regarding Claim 20,
Liao, Hunter, Smitherman, and Zadeh teach the entity information extraction from insurance card images of Claim 1 as described earlier.
Liao teaches,
including: a recurrent neural network (RNN)
(Liao [0390] ML model 3255 may include a neural network, such as one or more of a recurrent neural network (RNN)
for processing the tokens as character sequences along with spatial coordinates,
(Liao [Abstract] extracting contextually structured data
Liao [0022] identify semi-structured data generated by optical character recognition, the semi-structured data comprising a set of word tokens for each document image
Liao [0133] A set of text blocks may be generated around spatially separate portions text)
for integrating context from neighboring tokens based on spatial proximity,
(Liao [0134] such as by mapping OCR extractions to context (e.g., relating metadata to corresponding data).
Liao [0133] may utilize a frequency analysis, a set of proximate words, and/or metrics corresponding to the set of proximate words to identify metadata in a document image)
a hybrid convolutional neural network (CNN) for processing the image of the insurance card and combining an image representation with a hidden output from the RNN,
(Liao [0262] convolutional neural network (CNN) models. ... such as automatically contextualizing data extracted from an image.
Liao [0390] ML model 3255 may include a neural network, such as one or more of a recurrent neural network (RNN), a convolutional neural network (CNN), a region based CNN, or a Cascade region based CNN.
Liao [0248] In supervised training, each input in the training data is correlated to a desired output. This desired output may be a scalar, a vector, or a different type of data structure such as text or an image.
Liao [0258] The transformed output can be supplied to a subsequent layer, such as the hidden layer 1204, of the neural network 1200. The subsequent layer of the neural network 1200 can receive the transformed output, multiply the transformed output by a matrix of numeric weights and a nonlinearity, and provide the result to yet another layer of the neural network 1200. This process continues until the neural network 1200 outputs a final result at the output layer 120)
and a second RNN modeling the OCR output
(Liao [0009] the first text data comprises output from a first optical character recognition process and the second text data comprises output from a second optical character recognition process. In some embodiments, one or more of the first and second ML models comprise a recurrent neural network)
derived from the image of the insurance card.
(Liao [Abstract] extracting contextually structured data from document images, such as by automatically identifying document layout, document data, document metadata
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards)
Liao does not teach a graph neural network (GNN); using a fixed length feature vector
Hunter teaches,
a graph neural network (GNN)
(Hunter [0230] the deep neural network may be a deep convolutional neural network, deep graph neural network)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate the graph-manipulation based domain-specific execution teachings of Hunter “to generate a directed graph and determining a set of triggered vertices based on the directed graph” (Hunter [Abstract]). The modification would have been obvious, because it is merely applying a known technique (i.e. the graph-manipulation based domain-specific execution) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. determine “identifier to characterize a publisher of the event message. … a publisher may be an entity and may include various sources of an event message” Hunter [0056])
Hunter does not teach using a fixed length feature vector.
Collins teaches,
using a fixed length feature vector
( Collins [0440] can additionally include one or more fixed function or special function units to perform specific functions
Collins [0063] include specific values (e.g., [256] for a one-dimensional vector.)
It is prima facie obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the image content extraction of Liao to incorporate update multiple neural networks of Collins for “allow for multiple neural networks to be performed simultaneously and/or sequentially, and for results to be combined together” (Collins [0267]). The modification would have been obvious, because it is merely applying a known technique (i.e. update multiple neural networks) to a known concept (i.e. the image content extraction) ready for improvement to yield predictable result (i.e. “used to train and/or update neural networks based at least in part on input” Collins [0272])
Response to Remarks
Applicant's arguments filed on March 17, 2026, have been fully considered and Examiner’s remarks to Applicant’s amendments follow.
Response Remarks on Claim Objections
Applicant's amendments rectify the previous objections.
The objections are lifted.
Response Remarks on Claim Rejections - 35 USC § 112
Applicant's amendments rectify the previous rejections under 35 USC § 112.
The rejection under 35 USC § 112 is lifted.
Response Remarks on Claim Rejections - 35 USC § 101
The Applicant states:
“The claims do not recite agreements, contracts, legal obligations, advertising, marketing or sales activities or behaviors, or business relations. Instead, the claims recite a specific technical process for analyzing pixel-level image data through a pipeline of neural network architectures. Scoring a node of a fully connected neural network with a likelihood that the node corresponds to a member ID is an intra-computer technical operation that cannot reasonably be characterized as a commercial or legal interaction. Notably, the claims are directed to how the data is processed by specific computational architectures - not why it is processed or what business purpose it serves "
Examiner responds:
The “process for analyzing pixel-level image data through a pipeline … Scoring a node … with a likelihood that the node corresponds to a member ID … how the data is processed” amounts to gathering, sharing, and manipulation of data that expresses an Abstract Idea [Intellectual Ventures I v. Capital One Fin. Corp., 850 F.3d 1332, 121 USPQ2d 1940 (Fed. Cir. 2017) “collecting, displaying, and manipulating data” was considered part of the abstract idea], and Selecting A Particular Data Source or Type Of Data To Be Manipulated [Selecting information, based on types of information and availability of information in a power-grid environment, for collection, analysis and display, Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354-55, 119 USPQ2d 1739, 1742 (Fed. Cir. 2016)]
All the technical components (additional elements) function as designed with no unexpected results.
In the absence of unexpected results, changes or alteration of sequence do not make for a patentable invention, see Ex parte Rubin, 128 USPQ 440 (Bd. App. 1959) ; In re Burhans, 154 F.2d 690, 69 USPQ 330 (CCPA 1946); In re Gibson, 39 F.2d 975, 5 USPQ 230 (CCPA 1930)
The Applicant states:
“Here, the specific operations claimed … are plainly technical in nature. These operations define a specific machine-learning pipeline operating on image pixel data. That the ultimate output relates to insurance information does not convert these technical processing steps into a
method of organizing human activity any more than recognizing human faces in photographs constitutes organizing human activity (which it does not, per PEG Example 39). "
Examiner responds:
Examiner acknowledges many “technical” steps are recited, thusly they are deemed as additional element. However, all the “technical” components are generic in nature and perform as expected. For example, the Applicant’s Specification reads:
[0023] For capturing tokens from the insurance card 102 at OCR module 112, off-the-shelf or in-house optical character recognition systems can be employed.... Several different deep learning architectures can be used, including multimodal architectures which simultaneously process the tokens (character sequences) coming out of the OCR system
[0046] For all the above methods that use neural networks, includes RNNs, CNNs, GNNs, and the like, the training is performed via a gradient descent procedure … can be done through a variety of techniques including but not limited to random normal, Glorot normal, Glorot uniform, and He normal
[0049] While there exist a number of software libraries which assist in the training and deployment of neural networks, one software library that has been found to be advantageous is Tensorflow 2.0 (Google's neural network library), which contains the API needed to construct custom neural networks
[0050] in and out of the deep learning inference pipeline, can be defined as follows….from cloud storage (AWS S3) cloud storage (AWS S3) 314 are loaded into memory using the Tensorflow library
Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The additional elements merely add instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea, see MPEP 2106.05(f).
“[R]ecognizing human faces in photographs” does, in fact, constitute organizing human activity. Example 39 overcomes the abstract idea classification because it operates in a manner unlike generic machine learning algorithms and in a manner that is impossible for a human mind to perform. When a person identifies human faces, he or she cannot mentally apply and output “mirroring, rotating, smoothing, or contrast reduction” to a digital image presented to them. Therefore, the Example solved a problem in a manner unlike any done before and was not simply automating a manual process.
Applicant’s Specification provides uses “off-the-shelf” machine learning with no novel improvements upon utilized technical elements. That is, no additional element is utilized in a unique manner. Therefore, there is no improvement in the technology of machine learning.
The Applicant states:
“Further, Examiner's reliance on the specification's alleged reference to "off-the- shelf" OCR systems is misplaced. Id. at pg. 95-96. That individual components may be known does not render the specific combination obvious or generic. The inventive concept lies in the particular pipeline that solves a field-specific problem - not in any individual component. "
Examiner responds:
Applicant is correct that utilization of "off-the- shelf" components does not render an invention ineligible.
However, those generic components must either be utilized in a unique arrangement and yield an unexpected result.
Again, in the absence of unexpected results, changes or alteration of sequence do not make for a patentable invention, see Ex parte Rubin, 128 USPQ 440 (Bd. App. 1959) ; In re Burhans, 154 F.2d 690, 69 USPQ 330 (CCPA 1946); In re Gibson, 39 F.2d 975, 5 USPQ 230 (CCPA 1930)
Therefore, the rejection under 35 USC § 101 remains.
Response Remarks on Claim Rejections - 35 USC § 103
Applicant's amendments required the application of NO new/additional prior art.
The Applicant states:
“The Office's rejection now requires at least four, and at times five, references to sustain its position. While the number of cited references alone is not determinative, it is informative as to the weakness of the asserted combination…. Applicant respectfully maintains that several of the cited references are not analogous art…. not insurance card processing "
Examiner responds:
In response to Applicant's argument that the Examiner has combined an excessive number of references, reliance on a large number of references in a rejection does not, without more, weigh against the obviousness of the claimed invention. See In re Gorman, 933 F.2d 982, 18 USPQ2d 1885 (Fed. Cir. 1991).
All the references utilize machine learning for (character/object/text) recognition, extraction, and/or classification:
Liao [Abstract] Various embodiments are generally directed to techniques for image content extraction.
Hunter [0421] a message may encode a category label to determine a result based on vertices having the category label. For example, some embodiments may receive a message that includes the character “R” or the word “right” represent a category label associated with “rights” norm vertices.
Smitherman [0046] For example, the system can receive a string of characters such as “State Farm Property Casualty Payor, Inc.” and recognize “State Farm,” and match State Farm as a recognized payor in the system with specific rules on when, how, and who to bill.
Zedah [1477] The recognition of a pattern, color, person, face, logo, and text, including OCR (optical character recognition), is generally done by dissecting the image or video into pieces and components (including motion vectors for video, to track the objects, between the frames, as the difference between the neighboring frames) to find features or objects, and from the parameters associated with those features and objects
Collins [0267] a CNN executing on a DLA or a discrete GPU (e.g., GPU(s) 1420) may include text and word recognition, allowing reading and understanding
The Applicant states:
“The Office's rejection now requires at least four, and at times five, references to sustain its position. While the number of cited references alone is not determinative, it is informative as to the weakness of the asserted combination…. Applicant respectfully maintains that several of the cited references are not analogous art…. not insurance card processing "
Examiner responds:
Examiner considers “insurance card” as exemplary in nature; a form of intended use. The Specification describes:
[0002] Informational cards, such as credit cards, gift cards, insurances cards, and the like are widely used for a variety of purposes. In some circumstances, it can be advantages to extract information from these cards quickly and automatically.
[0003] Therefore there is a need for a system and method of processing cards, such as insurance cards, which accurately processes data about a card and adaptively changes based on feedback.
Effectively, intended use language is generally given limited patentable weight. See MPEP 2114(II) ("A claim containing a 'recitation with respect to the manner in which a claimed apparatus is intended to be employed does not differentiate the claimed apparatus from a prior art apparatus’ if the prior art apparatus teaches all the structural limitations of the claim. Ex parte Masham, 2 USPQ2d 1647 (Bd. Pat. App. & Inter. 1987).”); see also MPEP 2103(C).
The Examiner determines the aforementioned intended use statements do not result in any structural nor manipulative difference between the claimed invention and the prior art. Therefore, the intended use statements are afforded limited patentable weight.
Liao performs character recognition on a variety of documents and cards:
Liao [0297] in document images to identify metadata...insurance, government, disability claims, and medical records
Liao [0422] magnetic or optical cards
Liao [0041] identify semi-structured data generated by optical character recognition (OCR)
The Applicant states:
“This distinction is critical because member IDs on insurance cards are arbitrary alphanumeric strings (e.g., "XYZ123456789") that are not recognizable words. "
Examiner responds:
Liao considers arbitrary alphanumeric strings:
Liao [0141] may enable a computing device to digitize and/or extract key information from unstructured input data, such as document images.
Liao [0221] The event object may be created using a variety of formats including binary, alphanumeric, XML, etc. ...an event object may be a packed binary representation of a set of field values and include both metadata and field data associated with an event.
The Applicant states:
“Smitherman's "likelihood of a match" (para. [0079]) is a probability that a known piece of data matches a known patient record, not a probability that an unidentified character string corresponds to a particular type of information. "
Examiner responds:
Again, the application is rejected under a combination of references. The primary reference, Liao teaches:
Liao [0221] The event object may be created using a variety of formats including binary, alphanumeric, XML, etc. ...an event object may be a packed binary representation of a set of field values and include both metadata and field data associated with an event.
Even still, under the Applicant’s arguments, a name may represent a particular type of information:
Smitherman [0023] determining if a predetermined number of letters of a first name or a last name of the patient matches a predetermined number of letters of a first name or a last name of one of the plurality of candidate patients and if a date associated with the patient matches a date associated with the one of the plurality of candidate patients.
Smitherman [0045] The demographic information includes basic patient information, such as first name, last name, address, telephone number, date of birth, patient contact information, and claim data. The demographics information can also include a PAR (patient account record) record set.
For the remainder of the arguments: One cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In reMerck & Co., Inc., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). Where a rejection of a claim is based on two or more references, a reply that is limited to what a subset of the applied references teaches or fails to teach, or that fails to address the combined teaching of the applied references may be considered to be an argument that attacks the reference(s) individually. Where an applicant’s reply establishes that each of the applied references fails to teach a limitation and addresses the combined teachings and/or suggestions of the applied prior art, the reply as a whole does not attack the references individually as the phrase is used in Keller and reliance on Keller would not be appropriate. This is because "[T]he test for obviousness is what the combined teachings of the references would have suggested to [a PHOSITA]." In re Mouttet, 686 F.3d 1322, 1333, 103 USPQ2d 1219, 1226 (Fed. Cir. 2012).
Therefore, the rejection under 35 USC § 103 remains.
Prior Art Cited But Not Applied
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Jefferson ("INSURANCE VERIFICATION SYSTEM AND METHOD", U.S. Patent Number: US 6233563 B1, filed February 1999):
Jefferson [Col 2, Line 61 to Col 3, Line 7] FIGS. 1A and 1B illustrate front and back views, respectively, of an exemplary insurance card....As illustrated, the exemplary insurance card resembles a credit card. ...The back of the insurance card, as illustrated in FIG. 1B, includes a magnetic strip 10 and an area 15 in which the policyholder is to sign the card. The magnetic strip 10 contains all the information necessary to allow the nationwide database to verify that the cardholder has current vehicle insurance.
Jefferson [Col 5, Lines 8-15] vehicle insurance verification … Verification from the nationwide vehicle insurance database …, is accomplished by entering the identification (e.g., social security number or driver's license number) of each purchaser
thisvsthat.io (“Hashing vs. Tokenization”, submitted by Applicant) teaches Hashing and tokenization are both methods used to protect sensitive data, but they serve different purposes. Hashing involves converting data into a fixed-length string of characters, making it irreversible and secure for storage and comparison. On the other hand, tokenization replaces sensitive data with a randomly generated token, allowing the original data to be stored securely in a separate location. While hashing is ideal for securing passwords and sensitive information, tokenization is often used for payment processing and data masking in applications. Both methods are effective in enhancing data security and privacy.
Welling (“SYSTEMS AND METHODS FOR AUTOMATICALLY REDUCING DATA SEARCH SPACE AND IMPROVING DATA EXTRACTION ACCURACY USING KNOWN CONSTRAINTS IN A LAYOUT OF EXTRACTED DATA ELEMENTS”, U.S. Publication Number: 20110258195 A1) proposes automatically narrowing data search space and improving accuracy of data extraction using known constraints in a layout of extracted data elements for classified document.
Lucas (“MOBILE SUPPLEMENTATION, EXTRACTION, AND ANALYSIS OF HEALTH RECORDS”, U.S. Patent: US 10395772 B1) proposes capture, with a mobile device, a document such as a next generation sequencing (NGS) report that includes NGS medical information … information is extracted from the document … and the extracted information is provided into a structured data repository where it is accessible to provide information regarding the patient.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any extension fee pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to CHINEDU EKECHUKWU whose telephone number is (571)272-4493. The examiner can normally be reached on Mon-Fri 9 AM ET to 3:30 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Christine Behncke, can be reached on (571) 272-8103. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov.
Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/C.E./Examiner, Art Unit 3695
/CHRISTINE M Tran/Supervisory Patent Examiner, Art Unit 3695
Prosecution Insights