DETAILED ACTION
This action is made in response to the amendments/remarks filed on March 2, 2026. This action is made final.
Claims 1 and 4-20 are pending. Claims 2-3 have been previously cancelled. Claims 1, 17, and 18 have been amended. Claims 1 and 17 are independent claims.
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Applicant’s arguments with respect to the previous 101 rejection has been fully considered but are not persuasive.
Applicant argues the claims are not directed to a method of organizing human activity. However, the Examiner respectfully disagrees. MPEP 2106. 04(a)(2)(II) states that a claimed invention is directed to certain methods of organizing human activity if the identified claim elements contain limitations that encompass fundamental economic principles or practices, commercial or legal interactions, or managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions). The Examiner submits that the identified claim elements represent a series of rules or instructions that a person or persons, with or without the aid of a computer, would follow to standardize a patient health record from handwritten data. Furthermore, the Examiner submits that healthcare itself is inherently represents the organization of human activity. Insomuch as Applicant asserts limitation of the “pre-processing” the data record which includes “interpolating a portion of a text object into the data record” and using “machine learning-implement path tracing” as well using the “decoder layer” and “self-attention” layer of a “transformer-based machine learning model” are not directed to human activities, the examiner respectfully disagrees. Interpolating handwriting and/or using path tracing on the handwriting are steps a person can perform to improve the legibility of handwritten data. Furthermore, the use of a transformer-based machine learning model provides nothing more than mere instructions to implement the abstract idea on a generic computer (“apply it”), without placing any limitation on how to the model operates, supra. July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of item (c) at Pgs. 7-9. See MPEP 2106.05(f). MPEP 2106.05(f); July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of items (d) and (e) at Pgs. 8-9. Additionally, the transformer-based machine learning models having a decoder layer and self-attention layer, when given the broadest reasonable interpretation in light of the specification, the decoder layer and self-attention layer amounts to a mental process that creates data associations (e.g., see [0084] describing the decoder and attention mechanism as determining relationships on the data by applying different weights. As such, this decoder layer and self-attention layer are interpreted to be subsumed within the identified abstract idea, supra. July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of item (c) at Pgs. 7-9. See MPEP 2106.05(f). MPEP 2106.05(f); July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of items (d) and (e) and Example 48, claim 1, discussion of item (c). Because the claim elements fall under a series of rules or instructions that a person or persons would follow to convert handwritten documents into legible and standardized format, the claimed invention is directed to an abstract idea.
Applicant argues the amended claims, such as the pixel-level reconstruction, processing by a decoder transformer with multi-head self-attention, and generation of structured output and assignment of standardized activity codes using the trained model amount to significantly more. However, the examiner respectfully disagrees. MPEP 2106.05 outlines relevant considerations for evaluating whether additional elements amount to an inventive concept, such as improvement to the functioning of a computer, or to any other technology or technical field, applying the judicial exception with, or by use of a particular machine, effecting a transformation or reduction of a particular article to a different state or thing, applying or using the judicial exception in some other meaningful way beyond generally linking to a particular technological environment, or adding specific limitations other than what is well-understood, routing, or conventional. Applicant has failed to identify how any of the recited claim limitations amount to significantly more in accordance with MPEP 2106.05. As indicated in the 101 rejection below, the “pre-processing”, “interpolating” data, and “decoder” and “self-attention” layers were directed to the abstract idea, as those are steps a person can perform to process data. Insomuch as the data is processed at a “pixel-level”, merely amount to generally linking the abstract idea to a particular technological environment. (See MPEP 2106.04(d)(1) indicating generally linking an abstract idea to a particular technological environment does not amount to integrating the abstract idea into a practical application). Furthermore, the use of a machine learning model provides nothing more than mere instructions to implement the abstract idea on a generic computer (“apply it”), supra. July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of item (c) at Pgs. 7-9. See MPEP 2106.05(f). MPEP 2106.05(f); July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of items (d) and (e) at Pgs. 8-9. As such, the additional elements do not integrate the abstract idea into a practical application nor do they amount to an inventive concept and are rejected under 35 USC 101.
Applicant’s arguments with respect to the prior art rejection has been fully considered but is not persuasive. Insomuch as Applicant alleges Lelore fails to teach “reconstructing missing or discontinuous characters within a pixel-level representation of a scanned handwritten document”. The examiner respectfully disagrees. Applicant is reminded, one cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See In re Keller, 642 F.2d 413, 208 USPQ 871 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 231 USPQ 375 (Fed. Cir. 1986). In the present instance, Lucas teaches reconstruction of missing characters of a handwritten document, wherein the distinguishing characteristics of the document can identified on a pixel-level (e.g., see Fig. 5, [0049], [0053], [0054], [0057], [0092], [0093], [0127). As such, Lucas teaches “performing pixel-level interpolation of at least a portion of a text object within the image data, wherein interpolating comprises predictively reconstructing one or more missing discontinuous characters of the handwritten record to generated reconstructed image data”. While Lucas teaches machine learning models for reconstruction of missing characters of a handwritten document, Lucas fails to explicitly teach “path tracing”, which is taught by Lelore. Lelore teaches “machine learning-implemented path tracing of a handwritten record” (e.g., see [0093], [0105]-[0110] teaching the use of various machine learning techniques to aid in the recognition of handwritten input, including extracting a trajectory of the one or more strokes). Accordingly, it would have been obvious to modify Lucas in view of Lelore with a reasonable expectation of success. One would have been motivated to make the modification in order to improve upon optical character recognition often plagued with poor imaging that introduces spurious artefacts or noise onto the images by reconstructing the input based on where the handwriting could have traveled (e.g., see [0009], [0104] of Lelore). Accordingly, the combined teaching of Lucas in view of Lelore teach the claimed limitation.
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 and 4-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more.
Claims 1, 4-16, and 18-20 recite a method of assigning activity codes in a data record, which is within the statutory category of a process. Claim 17 recites a method of training a machine learning model, which is within the statutory class of a process.
Claims are eligible for patent protection under § 101 if they are in one of the four statutory categories and not directed to a judicial exception to patentability. Alice Corp. v. CLS Bank Int'l, 573 U.S. ___ (2014). Claims 1 and 4-20, each considered as a whole and as an ordered combination, are directed to a judicial exception (i.e., a law of nature, a natural phenomenon, or an abstract idea) without significantly more.
MPEP 2106 Step 2A – Prong 1:
The bolded limitations of:
Claim 1
identifying a first transformer-based machine learning model trained for conversion of data into a machine-readable format, wherein the first transformer-based machine learning model comprises an encoder-decoder architecture trained on handwritten image data and structured output data; identifying a second transformer-based machine learning model trained for assigning one or more predetermined activity codes to input data records, the second transformer-based machine learning model being distinct from the first transformer-based machine learning model, wherein the second transformer-based machine learning model is trained on structured input data; receiving at least one data record associated with a patient, the data record including one or more data items represented as image data, the image data comprising a digital scan of a handwritten record; performing, prior to tokenization, pre-processing the at least one data record to enhance legibility of at least one data item of the one or more data items of the at least one data record, wherein the pre-processing comprises performing pixel-level interpolation of at least a portion of a text object within the image data, wherein the interpolating comprises predictively reconstructing one or more missing or discontinuous characters of the handwritten record using machine learning-implemented path tracing of the handwritten record to generate reconstructed image data; using the first transformer-based machine learning model, converting at least a portion of the pre-processed at least one data record into at least one machine-readable data record, the converting at least the portion of the pre-processed at least one data record into at least one machine-readable data record comprising: processing the pre-processed at least one data record using a decoder layer of the first transformer-based machine learning model, the decoder layer comprising a multi-headed self-attention layer, configured to generate a context-aware decoding of the pre-processed at least one data record, which includes generating context-aware token embeddings representing the handwritten record; identifying a standardized record format; converting the machine-readable data record to the standardized record format; and using at least the second transformer-based machine learning model, assigning one or more predetermined activity codes to the at least one machine-readable data record in the standardized record format.
Claim 17
obtaining data representing a set of health records; applying one or more transformations to one or more of the health records to create a pre-processed set of health records, the transformations comprising: applying character recognition to at least one health record, classifying one or more portions of the at least one health record using a thresholding algorithm, and performing pixel-level interpolation of at least a portion of a text object within the image data, wherein the interpolating comprises predictively reconstructing one or more missing or discontinuous characters of the handwritten record using machine learning-implemented path tracing of the handwritten record to generate reconstructed image data prior to tokenization; creating a training set comprising the pre-processed set of health records; and training the model using the training set to convert at least a portion of the pre-processed at least one data record into at least one machine- readable data record, the converting at least the portion of the pre-processed at least one data record into at least one machine-readable data record comprising: processing the pre-processed at least one data record using a decoder layer of the first transformer-based machine learning model, the decoder layer comprising a self-attention layer, wherein the self-attention layer enables the decoder to generate a context-aware decoding of the pre-processed at least one data record.
as presently drafted, under the broadest reasonable interpretation, covers a method of organizing human activity (i.e., managing personal behavior including following rules or instructions) but for the recitation of generic computer components. For example, but for the noted computer elements, the claim encompasses a person following rules or instructions to process data in the manner described in the abstract idea, such as a healthcare worker receiving patient record image data and processing the documents to predict missing or discontinuous characters of the handwritten record prior to converting to a standardized format. The examiner further notes that “methods of organizing human activity” includes a person’s interaction with a computer (see October 2019 Update: Subject Matter Eligibility at Pg. 5). If the claim limitation, under its broadest reasonable interpretation, covers managing persona behavior or interactions between people but for the recitation of generic computer components, then it falls within the “method of organizing human activity” grouping of abstract ideas. Accordingly, the claim recites an abstract idea.
MPEP 2106 Step 2A – Prong 2:
This judicial exception is not integrated into a practical application because there are no meaningful limitations that transform the exception into a patent eligible application. The additional elements merely amount to instructions to apply the exception using generic computer components (“a processor”, "machine-readable format”, “machine-readable data”, and “digital records”—all recited at a high level of generality). Although they have and execute instructions to perform the abstract idea itself, this also does not serve to integrate the abstract idea into a practical application as it merely amounts to instructions to "apply it." (See MPEP 2106.04(d)(2) indicating mere instructions to apply an abstract idea does not amount to integrating the abstract idea into a practical application). Accordingly, the additional elements do not integrate the abstract idea into a practical application because they do not impose meaningful limits on practicing the abstract idea. Therefore, the claims are directed to an abstract idea.
The claims further recite “a first transformer-based machine-learning model, wherein the first transformer-based machine learning model comprises and encoder-decoder architecture trained on handwritten image data and structured output data”, “a second transformer-based machine learning model, wherein the second transformer-based machine learning model is trained on structured input data”. When given the broadest reasonable interpretation in light of the limited description of training in the disclosure, training of an machine-learning model with the noted data amounts to a mathematical concept or mental process that creates data associations. As such, this training of the machine-learning model is interpreted to be subsumed within the identified abstract idea and the use of the trained model provides nothing more than mere instructions to implement the abstract idea, supra. July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of item (c) at Pgs. 7-9. Furthermore, the use of a trained model provides nothing more than mere instructions to implement an abstract idea on a generic computer (“apply it”). See MPEP 2106.05(f). MPEP 2106.05(f); July 2024 Subject Matter Eligibility Examples, Example 47, Claim 2, discussion of items (d) and (e) at Pgs. 8-9.
The use of “optical character recognition” and “pixel-level” interpolation are not a generic computer component; however it is recited at a high levels of generality and similarly amount to generally linking the abstract idea to a particular technological environment. (See MPEP 2106.04(d)(1) indicating generally linking an abstract idea to a particular technological environment does not amount to integrating the abstract idea into a practical application).
The claims only manipulate abstract data elements as part of performing the abstract idea. They do not set forth improvements to another technological field or the functioning of the computer itself and instead use computer elements as tools in a conventional way to improve the functioning of the abstract idea identified above. Looking at the limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. There is no indication that the combination of elements improves the functioning of a computer or improves any other technology. Their collective functions merely provide conventional computer implementation. None of the additional elements recited "offers a meaningful limitation beyond generally linking 'the use of the [method] to a particular technological environment,' that is, implementation via computers." Alice Corp., slip op. at 16 (citing Bilski v. Kappos, 561 U.S. 610, 611 (U.S. 2010)).
At the levels of abstraction described above, the claims do not readily lend themselves to a finding that they are directed to a nonabstract idea. Therefore, the analysis proceeds to step 2B. See BASCOM Global Internet v. AT&T Mobility LLC, 827 F.3d 1341, 1349 (Fed. Cir. 2016) ("The Enfish claims, understood in light of their specific limitations, were unambiguously directed to an improvement in computer capabilities. Here, in contrast, the claims and their specific limitations do not readily lend themselves to a step-one finding that they are directed to a nonabstract idea. We therefore defer our consideration of the specific claim limitations’ narrowing effect for step two.") (citations omitted).
MPEP 2106 Step 2B:
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception for the same reasons as presented in Step 2A Prong 2. Moreover, the additional elements recited are known and conventional generic computing elements (“a processor”, "machine-readable format”, “machine-readable data”, and “digital records”—see Specification Fig. 6, [0018], [0065], [0075]-[0077] describing the various components as general purpose, common, standard, known to one of ordinary skill, and at a high level of generality, and in a manner that indicates that the additional elements are sufficiently well-known that the specification does not need to describe the particulars of such additional elements to satisfy the statutory disclosure requirements). Therefore, these additional elements amount to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept that amounts to significantly more. See MPEP 2106.05(f).
The Federal Circuit has recognized that "an invocation of already-available computers that are not themselves plausibly asserted to be an advance, for use in carrying out improved mathematical calculations, amounts to a recitation of what is 'well-understood, routine, [and] conventional.'" SAP Am., Inc. v. InvestPic, LLC, 890 F.3d 1016, 1023 (Fed. Cir. 2018) (alteration in original) (citing Mayo v. Prometheus, 566 U.S. 66, 73 (2012)). Apart from the instructions to implement the abstract idea, they only serve to perform well-understood functions (e.g., receiving, translating, and displaying data—see Specification above as well as Alice Corp.; Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307 (Fed. Cir. 2016); and Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334 (Fed. Cir. 2015) covering the well-known nature of these computer functions).
Similarly, as discussed above, use of “optical character recognition” and “pixel-level” interpolation were recited at a high level of generality and determined to generally link the abstract idea to a particular technological environment. This additional element has been re-evaluated under step 2B and has also been found insufficient to provide significantly more. (See MPEP 2106.05(A) indicating generally linking an abstract idea to a particular technological environment does not amount to significantly more).
Also, as discussed above with respect to integration of the abstract idea into a practical application, the additional elements using the transformer-based machine learning model were considered to be part of the abstract idea and “apply it,” respectively. This has been re-evaluated under the “significantly more” analysis and has also been found insufficient to provide significantly more. Regarding the training of the model is considered part of the abstract idea and thus cannot provide a practical application. Regarding the use of the model represented saying “apply it.” Using the trained model has been revaluated under the “significantly more” analysis and does not provide “significantly more” to the abstract idea. MPEP 2106.05(A) indicates also indicates that merely adding the words “apply it” or equivalent use cannot provide significantly more. Accordingly, even in combination, this additional element does not provide significantly more. As such the claim is not patent eligible.
Dependent Claims
The limitations of dependent but for those addressed below merely set forth further refinements of the abstract idea without changing the analysis already presented. Claims 4-9, 14 merely recite different pre-processing operations, claims 11, 12 merely recite the type of the data and machine readable format, claims 15 and 16 merely recite generating an output, and claims 18-20 merely recite making associations of the data to a particular category, label, or code, which covers a method of organizing human activity (i.e., managing personal behavior including following rules or instructions).
Claim 10 include the additional element of a “optical character recognition” that is analyzed in the same manner of the independent claim 17 and which does not provide a practical application or amounts to significantly more for the same reasons detailed above.
Claim 13 merely recite the use of an ensemble machine learning model that is analyzed in the same manner of the machine learning model of the independent claims and which does not provide a practical application or amounts to significantly more for the same reasons detailed above
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
Claim(s) 1, 4-16, and 18-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lucas et al. (USPPN:2020/0126663; hereinafter Lucas) in further view of Lelore et al. (USPPN: 2018/0285638; hereinafter Lelore), Sethumadhavan et al. (USPPN: 2018/0075192; hereinafter Sethumadhavan), and Vaswani et al. (“Attention Is All You Need”, [online], June 2017, retrieved from the internet: <URL: https://arxiv.org/abs/1706.03762>; hereinafter Vaswani).
As to claim 1, Lucas teaches A method, comprising:
Identifying a first machine learning model trained for conversion of data into a machine-readable format, wherein the machine learning model [is] trained on handwritten image data and structured output data (e.g., see [0049], [0056], [0066], [0088], [0100], [0119]-[0120] teaching one or more machine learning algorithms for converting data, including handwritten image data, to various machine-readable formats, wherein the machine learning model is trained using large collection of medical records to associate portions of the data to particular fields of a standard format);
receiving at least one data record associated with a patient, the data record including one or more data items represented as image data, the image data comprising a digital scan of a handwritten record (e.g., see Fig. 2, [0049], [0097] wherein the system received an image of a document having patient data, wherein the document includes scanned and handwritten documents);
performing, prior to tokenization, pre-processing, by the at least one processor, the at least one data record to enhance legibility of at least one data item of the one or more data items of the at least one data record, wherein the pre-processing comprises performing pixel-level interpolation of at least a portion of a text object within the image data, wherein the interpolating comprises predictively reconstructing one or more missing or discontinuous characters of the handwritten record using machine learning of the handwritten record to generate reconstructed image data (e.g., see Fig. 5, [0053], [0054], [0057], [0092], [0093], [0127] teaching the image may be pre-processed to perform text cleaning and error-detection, skew correction, etc. (i.e., enhance legibility) and can further include text cleaning and insertion of characters even when omitted using machine learning algorithms for prediction/estimate, wherein the document can be identified at a pixel level and pre-processing of the data is performed prior to any tokenization);
using the first machine learning model, converting, by the at least one processor, at least a portion of the pre-processed at least one data record into at least one machine-readable data record (e.g., see [0056], [0066] teaching the use of machine learning algorithms to convert the pre-processed data into a machine-readable structured data);
identifying, by the at least one processor, a standardized record format (e.g., see [0088], [0097], [0104]-[0108] teaching identifying a format and/or fields of the structured data);
converting the machine-readable data record to the standardized record format (e.g., see [0088], [0097], [0104]-[0108] wherein the structured data is converted in accordance with various format and/or field).
While Lucas teaches using machine learning to repair one or more records, including that of handwritten data, Lucas fails to explicitly teach using path tracing.
However, in the same field of endeavor character recognition, Lelore teaches machine learning-implemented path tracing of a handwritten record (e.g., see [0093], [0105]-[0110] teaching the use of various machine learning techniques to aid in the recognition of handwritten input, including extracting a trajectory of the one or more strokes). Accordingly, it would have been obvious to modify Lucas in view of Lelore with a reasonable expectation of success. One would have been motivated to make the modification in order to improve upon optical character recognition often plagued with poor imaging that introduces spurious artefacts or noise onto the images by reconstructing the input based on where the handwriting could have traveled (e.g., see [0009], [0104] of Lelore).
Lucas teaches utilizing at least one machine learning algorithm to extract health record data from an image and its conversion into a structured data repository where it can be further analyzed and/or accessed for medical billing information (e.g., see [0259]-[0261]). However, Lucas fails to explicitly teach identifying, by the at least one processor, a second machine learning model trained for assigning one or more predetermined activity codes to input data records; using at least the second machine learning model, assigning, by the at least one processor, one or more predetermined activity codes to the at least one machine-readable data record in the standardized record format.
In the same field of endeavor of healthcare data management, Sethumadhavan teaches identifying, by the at least one processor, a second machine learning model trained for assigning one or more predetermined activity codes to input data records, the second machine learning model being distinct from the first machine learning model, wherein the second machine learning model is trained on structured input data (e.g. see Figs. 5, 6, [0032], [0035], [0049], [0091], [0092] [0100] teaching a machine learning model trained to index/tag/map a code to input data records, wherein relevant structured data is associated with standardized codes during a training process); using at least the second machine learning model, assigning, by the at least one processor, one or more predetermined activity codes to the at least one machine-readable data record in the standardized record format (e.g., see Figs. 5, 6, [0032], [0035], [0076], [0092], [0100] teaching indexing/tagging/mapping a code to the inputted medical document, the index/tag/mapping including actionable codes and/or billing information, which is consistent with at least [0015] of Applicant’s disclosure of an “activity code”).
Accordingly, it would have been obvious to modify the automatic converting and labeling of medical data taught in Lucas-Lelore with the specific medical coding of Sethumadhavan to automatically and efficiently code medical records thereby improving revenue and treatment efficacy (e.g., see [0008]-[0009] of Sethumadhavan).
While Lucas, Lelore, and Sethumadhaven teach machine learning models, Lucas-Lelore-Sethumadhaven fail to teach transformer-based machine learning models, wherein the transformer-based machine learning model comprises an encoder-decoder architecture; a second transformer-based machine learning model; the converting at least the portion of the pre-processed at least one data record into at least one machine-readable data record comprising: processing the pre-processed at least one data record using a decoder layer of the first transformer-based machine learning model, the decoder layer comprising a multi-headed self-attention layer, configured to generate a context-aware decoding of the pre-processed at least one data record, which includes generating context-aware token embeddings representing the record.
However, in the same field of endeavor of machine learning models for classifying a plurality of data sets, Vaswani teaches transformer-based machine learning model comprises an encoder-decoder architecture; a second transformer-based machine learning model; the converting at least the portion of the pre-processed at least one data record into at least one machine-readable data record comprising: processing the pre-processed at least one data record using a decoder layer of the first transformer-based machine learning model, the decoder layer comprising a multi-headed self-attention layer, configured to generate a context-aware decoding of the pre-processed at least one data record, which includes generating context-aware token embeddings representing the record (e.g., see Abstract, Section 3: “Model Architecture, “Encoder and Decoder Stacks”, “Multi-Head Attention”, “Embeddings and Softmax”, “Positional Encoding” teaching the use of one transformer-based machine learning models for processing data, which uses one or more encoders and/or decoders having a multi-head attention layer for providing context aware embedded tokens).
Accordingly, it would have been obvious to modify Lucas-Lelore-Sethumadhaven in view of Vaswani with a reasonable expectation of success. One would have been motivated to make the modification in order as a simple substitution of one known type of machine learning model for another, to yield the predictable results of an improved model that requires less time to train (e.g., See KSR Int’l v. Teleflex Inc., 127 S. Ct. 1727, 1740-41, 82 USPQ2d 1385, 1396 (2007); and MPEP 2143; Abstract of Vaswani).
As to claim 4, the rejection of claim 1 is incorporated. Lucas further teaches wherein the interpolating further comprises using machine learning-implemented path tracing of the handwritten record to increase legibility of one or more characters of the handwritten record, or insert one or more characters into the handwritten record (e.g., see [0054], [0093] wherein the pre-processing includes text cleaning and insertion of characters).
As to claim 5, the rejection of claim 1 is incorporated. Lucas further teaches wherein the pre-processing comprises at least one of: rotating the data record, rotating a text object of the data record, removing a visual artifact of the data record, adjusting a brightness, optical curve, or contrast of the data record, changing a bit depth of image data of the data record, or superimposing a visual aid onto image data of the data record (e.g., see Fig. 3, [0053], [0078], [0189] teaching processing of the image includes rotating/skewing the image, removing irregularities, resolution conversion, generating a mask to outline the image, etc.).
As to claim 6, the rejection of claim 5 is incorporated. Lucas further teaches wherein rotating a text object of the data record is incorporated into a process for parallelizing a plurality of text objects of the data record (e.g., see [0189] teaching rotating/skewing the image to line it up with a template (i.e., parallelizing)).
As to claim 7, the rejection of claim 5 is incorporated. Lucas further teaches wherein the visual artifact is a scanned dust speck or scanned print error (e.g., see also [0189] teaching the removal of image irregularities. It is further noted that the type of irregularity is interpreted as a matter or intended use as it imparts no structural difference to the claimed invention. An intended use generally does not impart a patentable distinction if it merely states an intention or is a description of how the claimed apparatus is to be used. (See MPEP 2111.05)).
As to claim 8, the rejection of claim 5 is incorporated. Lucas further teaches wherein the visual aid is a bounding box (e.g., see Fig. 2, [0078] wherein the visual aid is a bounded box).
As to claim 9, the rejection of claim 8 is incorporated. Lucas further teaches wherein converting at least the portion of the pre-processed data record comprises assigning a text object of the pre-processed data record to a field, wherein the field is based at least in part on an identifier of the bounding box (e.g., see Fig. 2, [0078]-[0079] wherein the text is assigned to a particular field based on a matching string of the bounded box).
As to claim 10, the rejection of claim 1 is incorporated. Lucas further teaches wherein the converting at least the portion of the pre-processed data record into a machine-readable format is performed using optical character recognition (e.g., see [0066] teaching OCR techniques to convert the data to a structured format).
As to claim 11, the rejection of claim 1 is incorporated. Lucas further teach wherein the standardized record format is Health Level 7 (HL7) Fast Healthcare Interoperability Resources (FHIR) (It is noted that the type of format is interpreted as a matter or intended use as it imparts no structural difference to the claimed invention. An intended use generally does not impart a patentable distinction if it merely states an intention or is a description of how the claimed apparatus is to be used. (See MPEP 2111.05). Nonetheless, see [0262] teaching the structured data format includes FHIR).
As to claim 12, the rejection of claim 11 is incorporated. Lucas further teaches further comprising converting the machine-readable data record into a different version of HL7 (It is noted that the type of format is interpreted as a matter or intended use as it imparts no structural difference to the claimed invention. An intended use generally does not impart a patentable distinction if it merely states an intention or is a description of how the claimed apparatus is to be used. (See MPEP 2111.05). Nonetheless, see [0262] teaching the structured data format includes any model that supports the representation of numerous attributes which can them be used in a health system’s EHR (i.e., different version of HL7)).
As to claim 13, the rejection of claim 1 is incorporated. Lucas teaches converting at least the portion of the pre-processed data record into a machine-readable format using a machine learning model (e.g., see rejection of claim 1 above), Lucas-Lelore-Sethumadhavan fail to teach further teaches using an ensemble machine learning model.
However, in the same field of endeavor machine learning models for classifying data, Vaswami teaches using an ensemble machine learning model (e.g., see Abstract teaching ensemble-based models).
Accordingly, it would have been obvious to modify the machine learning models of Lucas-Lelore-Sethumadhavan with the ensemble machine learning model of Vaswami with a reasonable expectation of success. One would have been motivated to make the modification as a simple substitution of one known method for another to yield the predictable results of minimizing errors functions (e.g., See KSR Int’l v. Teleflex Inc., 127 S. Ct. 1727, 1740-41, 82 USPQ2d 1385, 1396 (2007); and MPEP 2143).
As to claim 14, the rejection of claim 1 is incorporated. Lucas further teaches wherein converting at least the portion of the pre-processed data record comprises performing a spelling check or a grammar check (e.g., see [0115] wherein error correction includes spell checking or context-based correlation).
As to claim 15, the rejection of claim 1 is incorporated. Lucas-Lelore fail to teach generating an electronic report comprising an algorithmically-generated explanation of the assigning of the activity codes.
In the same field of endeavor of healthcare data management, Sethumadhavan teaches generating an electronic report comprising an algorithmically-generated explanation of the assigning of the activity codes (e.g., see [0088] wherein a result is generating comprising an algorithmically-generated explanation of the code assignment).
Accordingly, it would have been obvious to modify the automatic converting and labeling of medical data taught in Lucas-Lelore with the specific medical coding of Sethumadhavan to automatically and efficiently code medical records thereby improving revenue and treatment efficacy (e.g., see [0008]-[0009] of Sethumadhavan).
As to claim 16, the rejection of claim 1 is incorporated. Lucas further teaches further comprising generating an electronic claim file from the machine-readable data record (e.g., see [0271] wherein the data is for validating a claim with patient’s insurance).
As to claim 18, the rejection of claim 1 is incorporated. Lucas further teaches wherein the first transformer-based machine learning model is trained to associate portions of digitized text of a digitized record with particular categories or labels associated with one or more fields of the standardized format (e.g., see [0088], [0124], [0189] wherein a classification model is trained to use the predetermined region locations of the standardized format to extract key health information).
While Lucas teaches machine learning models, Lucas fails to teach transformer-based machine learning models
However, in the same field of endeavor of machine learning models for classifying a plurality of data sets, Vaswami teaches transformer-based machine learning models (e.g., see Abstract teaching the use of transformer-based machine learning models).
Accordingly, it would have been obvious to modify Lucas-Lelore-Sethumadhaven in view of Vaswami with a reasonable expectation of success. One would have been motivated to make the modification in order as a simple substitution of one known type of machine learning model for another, to yield the predictable results of an improved model that requires less time to train (e.g., See KSR Int’l v. Teleflex Inc., 127 S. Ct. 1727, 1740-41, 82 USPQ2d 1385, 1396 (2007); and MPEP 2143; Abstract of Vaswani).
As to claim 19, the rejection of claim 1 is incorporated. Lucas fails to teach wherein assigning one or more predetermined activity codes to the at least one machine-readable data record in the standardized record format is performed at least in part by associating at least a portion of text relating to one or more fields of the standardized format record with at least one activity code of the one or more predetermined activity codes.
However, in the same field of endeavor of healthcare data management, Sethumadhavan teaches wherein assigning one or more predetermined activity codes to the at least one machine-readable data record in the standardized record format is performed at least in part by associating at least a portion of text relating to one or more fields of the standardized format record with at least one activity code of the one or more predetermined activity codes (e.g., see Fig. 9, [0084], [0090]-[0092], [0097] wherein various text is associated with code based on a belief network using previous records). Accordingly, it would have been obvious to modify the automatic converting and labeling of medical data taught in Lucas-Lelore with the specific medical coding of Sethumadhavan to automatically and efficiently code medical records thereby improving revenue and treatment efficacy (e.g., see [0008]-[0009] of Sethumadhavan).
As to claim 20, the rejection of claim 18 is incorporated. While Lucas teach machine learning models for making data associations (see rejection of claim 18 above), Lucas-Lelore-Sethumadhaven fail to teach wherein the associating the portions of the digitized text of the digitized record with the particular categories or labels is performed at least in part by using the self-attention layer to capture a relative significance and relationship amongst different portions or patches of the image data
However, in the same field of endeavor of machine learning models for classifying a plurality of data sets, Vaswami teaches wherein the associating the portions of the digitized text of the digitized record with the particular categories or labels is performed at least in part by using the self-attention layer to capture a relative significance and relationship amongst different portions or patches of the image data (e.g., see “Background”, “Attention”, “Why Self-Attention” teaching the use of a self-attention layer for making data associations by capturing dependencies of different positions of the input data).
Accordingly, it would have been obvious to modify Lucas-Lelore-Sethumadhaven in view of Vaswami with a reasonable expectation of success. One would have been motivated to make the modification in order as a simple substitution of one known type of machine learning model for another, to yield the predictable results of an improved model that requires less time to train (e.g., See KSR Int’l v. Teleflex Inc., 127 S. Ct. 1727, 1740-41, 82 USPQ2d 1385, 1396 (2007); and MPEP 2143; Abstract of Vaswani).
Claim(s) 17 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lucas et al. (USPPN:2020/0126663; hereinafter Lucas) in further view of Lelore et al. (USPPN: 2018/0285638; hereinafter Lelore) and Vaswani et al. (“Attention Is All You Need”, [online], June 2017, retrieved from the internet: <URL: https://arxiv.org/abs/1706.03762>; hereinafter Vaswani).
As to claim 17, Lucas teaches A computer-implemented method of training a machine learning model (e.g., see [0228], [0273] teaching training a machine learning algorithm), comprising:
obtaining, by the at least one processor, data representing a set of digital health records (e.g., see [0228], [0273] wherein medical information is received);
applying, by the at least one processor, one or more transformations to one or more of the digital health records to create a pre-processed set of digital health records, the transformations comprising: applying optical character recognition to at least one digital health record, classifying one or more portions of the at least one digital health record using a thresholding algorithm, and performing pixel-level interpolation of one or more characters identified in the at least one digital health record, wherein the interpolation comprises predictively reconstructing one or more missing or discontinuous characters of the handwritten record using machine learning of the handwritten record to generate reconstructed image data prior to tokenization (e.g., see Fig. 5, [0053], [0054], [0057], [0092], [0093], [0127] teaching the image may be pre-processed, including OCR, to perform text cleaning and error-detection, skew correction, etc. (i.e., enhance legibility) and can further include text cleaning and insertion of characters even when omitted using machine learning algorithms for prediction/estimate, wherein the document can be identified at a pixel level and pre-processing of the data is performed prior to any tokenization);
creating a training set comprising the pre-processed set of digital health records (e.g., see [0119], [0229] wherein medical/clinical text is based on OCR text cleaning); and
training the machine learning model using the training set, to convert at least a portion of the pre-processed at least one data record into at least one machine-readable data record, the converting at least the portion of the pre-processed at least one record into at least one machine-readable data (e.g., see [0088], [0097], [0104]-[0108] wherein the structured data is converted in accordance with various format and/or field (e.g., see [0229] teaching a training feedback loop).
Lucas teaches rotating or interpolating one or more characters identified in the at least one digital health record including using machine-learning of a handwritten record of the digital health record to generate reconstructed image data (e.g., see [0054], [0093] wherein the pre-processing includes text cleaning and insertion of characters).
While Lucas teaches using machine learning to repair one or more records, Lucas fails to explicitly teach using path tracing.
However, in the same field of endeavor character recognition, Lelore teaches machine learning-implemented path tracing of a handwritten record to generate reconstructed image data (e.g., see [0093], [0105]-[0110] teaching the use of various machine learning techniques to aid in the recognition of handwritten input, including extracting a trajectory of the one or more strokes). Accordingly, it would have been obvious to modify Lucas in view of Lelore with a reasonable expectation of success. One would have been motivated to make the modification in order to improve upon optical character recognition often plagued with poor imaging that introduces spurious artefacts or noise onto the images by reconstructing the input based on where the handwriting could have traveled (e.g., see [0009], [0104] of Lelore).
While Lucas and Lelore teach machine learning models, Lucas-Lelore-Sethumadhaven fail to teach transformer-based machine learning models, the converting at least the portion of the pre-processed at least one data record into at least one machine-readable data record comprising: processing the pre-processed at least one data record using a decoder layer of the transformer-based machine learning model, the decoder layer comprising a multi-headed self-attention layer, configured to generate a context-aware decoding of the pre-processed at least one data record, which includes generating context-aware token embeddings representing the record.
However, in the same field of endeavor of machine learning models for classifying a plurality of data sets, Vaswani teaches transformer-based machine learning model the converting at least the portion of the pre-processed at least one data record into at least one machine-readable data record comprising: processing the pre-processed at least one data record using a decoder layer of the transformer-based machine learning model, the decoder layer comprising a multi-headed self-attention layer, configured to generate a context-aware decoding of the pre-processed at least one data record, which includes generating context-aware token embeddings representing the record (e.g., see Abstract, Section 3: “Model Architecture, “Encoder and Decoder Stacks”, “Multi-Head Attention”, “Embeddings and Softmax”, “Positional Encoding” teaching the use of one transformer-based machine learning models for processing data, which uses one or more encoders and/or decoders having a multi-head attention layer for providing context aware embedded tokens).
Accordingly, it would have been obvious to modify Lucas-Lelore-Sethumadhaven in view of Vaswani with a reasonable expectation of success. One would have been motivated to make the modification in order as a simple substitution of one known type of machine learning model for another, to yield the predictable results of an improved model that requires less time to train (e.g., See KSR Int’l v. Teleflex Inc., 127 S. Ct. 1727, 1740-41, 82 USPQ2d 1385, 1396 (2007); and MPEP 2143; Abstract of Vaswani).
Relevant Art not Cited
As a courtesy, the following prior art documents have been found during the course of examination and deemed relevant to applicant’s disclosure. Applicant is strongly encouraged to review the following references prior to any amendments/remarks:
Molenda (USPPN: 2023/0368878): Systems and methods using multidimensional language and vision models and maps to categorize, describe, coordinate, and track anatomy and health data
Madan (USPPN: 2011/0040576): Converting arbitrary text to formal medical code
Schneider et al. (USPPN: 2018/0011974): Systems and methods for improved optical character recognition of health records
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. “The use of patents as references is not limited to what the patentees describe as their own inventions or to the problems with which they are concerned. They are part of the literature of the art, relevant for all they contain.” In re Heck, 699 F.2d 1331, 1332-33, 216 USPQ 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 USPQ 275, 277 (CCPA 1968)). Further, a reference may be relied upon for all that it would have reasonably suggested to one having ordinary skill the art, including nonpreferred embodiments. Merck & Co. v. Biocraft Laboratories, 874 F.2d 804, 10 USPQ2d 1843 (Fed. Cir.), cert. denied, 493 U.S. 975 (1989). See also Upsher-Smith Labs. v. Pamlab, LLC, 412 F.3d 1319, 1323, 75 USPQ2d 1213, 1215 (Fed. Cir. 2005); Celeritas Technologies Ltd. v. Rockwell International Corp., 150 F.3d 1354, 1361, 47 USPQ2d 1516, 1522-23 (Fed. Cir. 1998).
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 nonprovisional extension fee (37 CFR 1.17(a)) 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 mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to STELLA HIGGS whose telephone number is (571)270-5891. The examiner can normally be reached Monday-Friday: 9-5PM.
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, Peter Choi can be reached on (469) 295-9171. 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.
/STELLA HIGGS/Primary Examiner, Art Unit 3686
Prosecution Insights