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 .
Status of the Claims
The status of the claims as of the response filed 5/21/2026 is as follows: Claims 4, 6, and 10 are cancelled, and all previously given rejections for these claims are considered moot. Claims 1-3, 5, 7-9, and 11 are currently amended. Claims 1-3, 5, 7-9, and 11 are currently pending in the application and have been considered below.
Note: Applicant is reminded of the requirements of 37 CFR 1.121 as detailed in MPEP 714(II)(C)(B): “All claims being currently amended must be presented with markings to indicate the changes that have been made relative to the immediate prior version. The changes in any amended claim must be shown by strike-through (for deleted matter) or underlining (for added matter).” Examiner notes that at least claim 11 does not comply with this requirement (it includes “non-transitory” as an underlined element, indicating that this is newly-introduced in the 5/21/2026 amendment even though these words were already present in the immediate prior version of the preliminary amendment filed 1/3/2025) and requests that all future amendments comply with this requirement.
Response to Amendment
Interpretation Under 35 USC 112(f)
Claim 1 has been amended such that it no longer invokes 35 USC 112(f) interpretation. The outstanding 35 USC 112(a) and (b) rejections stemming from the 112(f) interpretation are thus withdrawn.
Rejection Under 35 USC 112(b)
Claims 1 and 7 have been amended to sufficiently clarify the indefinite elements and limitations such that the corresponding 35 USC 112(b) rejections are withdrawn.
Response to Arguments
Rejection Under 35 USC 101
On pages 10-12 of the response filed 5/21/2026 Applicant argues that the amended independent claims include “an additional feature or a combination of features which demonstrates an improvement to the conventional computer-implemented diagnostic modeling technology.” Applicant specifically asserts that “the claimed invention addresses the technical problem of loss of critical features of during data compression and lower predictive performance associated with the conventional computer-implemented diagnostic modeling technology” because “it recites a specific technical architecture – an autoencoder with a bottleneck layer integrated with a classification layer – that is optimized via a unique weighted dual-loss function” which allows for “simultaneously minimizing both the reconstruction error, while ensuring the compressed data still looks like the original, and the prediction error, while ensuring the compressed data is useful for diagnosis,” which “clearly improves the conventional computer-implemented diagnostic modeling technology.” Applicant concludes that such improvements are demonstrated in the experimental results illustrated in Fig. 5, which shows the claimed model’s performance compared to other conventional model types.
Applicant’s arguments are fully considered, but are not persuasive. Examiner notes that the compression and optimization functions performed by the various layers/components of the neural network are mathematical concepts such that they are part of the abstract idea itself, and thus they are not evaluated as additional elements under Steps 2A – Prong 2 and 2B. Improvements to the abstract idea itself (e.g. the mathematical processing and representations of data to perform predictive medical diagnosis) are not considered technical improvements to a computer or other technical field, and do not provide integration into a practical application (see MPEP 2106.05(a)(II): “it is important to keep in mind that an improvement in the abstract idea itself… is not an improvement in technology”). Further, Applicant’s specification does not lay out specific technical problems with existing or conventional machine learning model architectures that this invention seeks to solve; the issues that the invention appears to seek to solve instead include drawbacks with conventional human workflows in clinician-led diagnosis of certain conditions like stomach cancer (see paras. [04]-[05]), to which known deep learning methodologies are applied (see paras. [02]-[03]). Though para. [76] describes alleged performance improvements of the model against other types of reference models, it does not link such improvements to specific aspects of the model architecture or training methodology, or technical issues with the reference model methodologies.
On pages 12-13 Applicant argues that “the use of a final cost function as a linear sum with individual weights applied to reconstruction and prediction errors represents a non-conventional optimization strategy” and thus confers eligibility under Step 2B. Applicant’s arguments are fully considered, but are not persuasive. As indicated above, the dual optimization of the autoencoder and classification layer via a weighted linear sum of two cost functions describes mathematical concepts that are part of the abstract idea itself. Accordingly, the conventionality of this feature is not evaluated under Step 2B, because it is not an additional element beyond the abstract idea itself; even unconventional abstract ideas/functions are still abstract and do not confer eligibility.
For the reasons outlined above, the 35 USC 101 rejections are upheld.
Rejection Under 35 USC 103
On pages 14-16 Applicant argues that the combination of Irving and Vasiljeva fails to teach optimizing the autoencoder and the classification layer simultaneously to minimize a final cost function as in the amended independent claims. Applicant specifically asserts that though Irving “mentions calculating both reconstruction and classification errors, it describes the combination of these losses as optional or refers generally to an overall loss” such that it “does not teach or suggest the critical requirement of simultaneously updating both the autoencoder and classification components using a weighted linear sum to tune the latent space at the bottleneck layer” (emphasis original). Applicant’s arguments are fully considered, but are not persuasive. Examiner maintains that Irving sufficiently suggests this feature of the amended claims. In paras. [0056]-[0057] the system calculates and optionally combines reconstruction error of the autoencoder and classification error of the classification layer via a linear weighted sum equation that applies individual weights to each of the reconstruction loss and the classification loss to calculate an overall loss L. Just because this function is “optional” in the system of Irving does not mean it is not taught for the purposes of prior art; Examiner notes that Applicant’s own specification has similar optional language with regard to the calculation of a final cost, with para. [15] stating “the cost function application unit may apply a final cost function as a linear sum of a first cost function that calculates a reconstruction error of the autoencoder and a second cost function that calculates a prediction error of the classification layer, and apply individual weights to the first cost function and the second cost function to apply the final cost function” and para. [64] stating “the cost function application unit 143 may apply a final cost function Lss, which is a linear sum of a first cost function Lmse and a second cost function Lbce” (emphasis added). This linear weighted sum of reconstruction error and classification error is disclosed in paras. [0056]-[0057] of Irving, and it is irrelevant that such a feature is described as optional because it is still disclosed as a function of the system. Paras. [0005], [0010], & [0040] of Irving then show that the loss function L (e.g. the combined loss function resulting from the linear weighted sum as in [0056]-[0057]) is used to train the overall neural network architecture, which includes both an autoencoder and a classification layer as shown in Fig. 3B & at least para. [0004]. Because the entire neural network architecture may be trained/optimized using the combined loss function, the autoencoder and classification layers are considered to be “simultaneously” optimized as required by the amended claim language.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 5/21/2026 is in accordance with the provisions of 37 CFR 1.97 and is considered by the Examiner.
Specification
The disclosure is objected to because of the following informalities: the descriptions of Figs. 4 and 5 in paras. [27]-[28] appear to be erroneously switched; Fig. 4 depicts the flowchart described in [28] while Fig. 5 depicts the graph described in [27]. Appropriate correction is required.
Claim Objections
Claim 11 is objected to because of the following informalities: it recites “a set of instruction” which should be amended to “a set of instructions” for improved grammatical clarity. Appropriate correction is required.
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 1-3, 5, 7-9, and 11 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1
In the instant case, claims 1-3 and 5 are directed to a device (i.e. a machine), claims 7-9 are directed to a method (i.e. a process), and claim 11 is directed to a computer-readable program stored on a non-transitory computer-readable storage medium (i.e. a manufacture). Thus, each of the claims falls within one of the four statutory categories. Nevertheless, the claims fall within the judicial exception of an abstract idea.
Step 2A – Prong 1
Independent claims 1 and 7 recite steps that, under their broadest reasonable interpretations, describe a mental process and mathematical concepts. Specifically, claim 1 (as representative) recites:
at least one processor; and
a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to:
collect a medical diagnosis data of a plurality of patients;
generate input data by embedding the medical diagnosis data of the plurality of patients into a binary vector format;
generate a deep neural network-based disease-of-interest prediction model comprising an autoencoder having a bottleneck layer and a classification layer connected to the bottleneck layer;
train the deep neural network-based disease-of-interest prediction model by:
(i) inputting the input data into the autoencoder to generate compressed data at the bottleneck layer and reconstructing the input data based on the compressed data;
(ii) predicting a development of disease of interest via the classification layer based on the compressed data; and
(iii) optimizing the autoencoder and the classification layer simultaneously to minimize a final cost function, wherein the final cost function is a linear sum of a first cost function calculating a reconstruction error of the autoencoder and a second cost function calculating a prediction error of the classification layer, the linear sum applying individual weights to the first cost function and the second cost function,
wherein a future development of disease of interest of a patient is predicted using the generated deep neural-network disease-of-interest prediction model based on existing medical diagnosis data of the patient.
But for the recitation of generic computer components like a processor, memory, and deep neural network prediction model, the italicized functions, when considered as a whole, describe mathematical concepts and mental processes for fitting and using a predictive model to make medical diagnoses. For example, a clinician, researcher, or other knowledgeable person could, either mentally or with the aid of pen and paper, collect medical diagnosis data of multiple patients, generate binary vector embeddings of the diagnosis data, fit a predictive model (e.g. one “based” on deep neural network concepts with various layers) to the embeddings to predict a development of a disease of interest, and use the generated model to predict when a patient may develop a disease. The method of fitting/training the DNN-based model that includes inputting data into an autoencoder to generate compressed data and reconstruct the input data and optimizing the autoencoder and classification layers to minimize a final cost function via a weighted linear sum of two cost functions amount to mathematical concepts because such functions are performed via mathematical compression and decompression calculations and linear summation and function minimization calculations to arrive at a mathematically-represented predictive model. Accordingly, claim 1 recites an abstract idea in the form of a mental process and mathematical concepts. Claim 7 recites substantially similar subject matter as claim 1 and is found to recite an abstract idea under the same analysis.
Dependent claims 2-3, 5, 8-9, and 11 inherit the limitations that recite an abstract idea from their dependence on claims 1 or 7, and thus these claims also recite an abstract idea under the Step 2A – Prong 1 analysis. In addition, claims 2-3, 5, and 8-9 recite additional limitations that further describe the abstract idea identified in the independent claims. Specifically, claims 2-3 and 8-9 describe extracting and counting ICD codes, then vectorizing them into a binary vector of the counted size. A human actor could achieve these functions either mentally or with aid of pen and paper by observing medical records, identifying ICD codes for different patients, and constructing binary vector representations corresponding to the number of ICD codes observed. Claim 5 further describe the mathematical layers and functions of the prediction model, which further describe the mathematical concepts introduced by the independent claims and are thus also abstract.
However, recitation of an abstract idea is not the end of the analysis. Each of the claims must be analyzed for additional elements that indicate the abstract idea is integrated into a practical application to determine whether the claim is considered to be “directed to” an abstract idea.
Step 2A – Prong 2
The judicial exception is not integrated into a practical application. In particular, independent claims 1 and 7 do not include additional elements that integrate the abstract idea into a practical application. The additional elements of claims 1 and 7 include a processor executing instructions stored in a memory to implement the various functions and deep neural network-based disease-of-interest prediction model. These additional elements, when considered in the context of each claim as a whole, merely serve to automate the mental and mathematical operations of the claims, and thus amount to instructions to “apply” the abstract idea using generic computer components (see MPEP 2106.05(f)). In other words, the otherwise-abstract mathematical and mentally-based functions of the claims being performed via the recited processors merely digitizes and/or automates the functions such that they occur in a computerized environment. Accordingly, claims 1 and 7 as a whole are each directed to an abstract idea without integration into a practical application.
The judicial exception recited in dependent claims 2-3, 5, 8-9, and 11 is also not integrated into a practical application under a similar analysis as above. Claims 2-3, 5, and 8-9 are performed with the same additional elements introduced in the independent claims, without introducing any new additional elements of their own, and accordingly also amount to mere instructions to apply the abstract idea with these same additional elements. Claim 11 recites a non-transitory computer-readable recording medium storing a set of instructions that cause one or more processor of a deep neural network-based disease-of-interest prediction device to perform the method of claim 7, which also amounts to instructions to “apply” the exception as explained for the independent claims above.
Accordingly, the additional elements of claims 1-3, 5, 7-9, and 11 do not integrate the abstract idea into a practical application because they do not impose any meaningful limits on practicing the abstract idea. Claims 1-3, 5, 7-9, and 11 are directed to an abstract idea.
Step 2B
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of a processor and memory for performing the collecting, generating, training, inputting, predicting, optimizing, etc. steps of the invention amount to mere instructions to apply the exception using generic computer components. As evidence of the generic nature of the above recited additional elements, Examiner notes para. [85] of Applicant’s specification, noting various known computer memory media and stating that “functional programs, codes, and code segments for implementing this embodiment will be easily understood by those skilled in the art.” This disclosure does not indicate that the elements of the invention are particular machines, and instead provide generic examples of computer hardware such that one of ordinary skill in the art would understand that any generic computing units executing appropriate code could be utilized to implement the invention.
Analyzing these additional elements as an ordered combination adds nothing that is not already present when considering the elements individually; the overall effect of the processor and memory is to digitize and/or automate mathematical operations and mental processes for fitting and using a predictive model to make medical diagnoses. Thus, when considered as a whole and in combination, claims 1-3, 5, 7-9, and 11 are not patent eligible.
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 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.
The factual inquiries 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 1-3, 5, 7-9, and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Irving et al. (US 20230154627 A1) in view of Vasiljeva et al. (Reference U on the PTO-892 mailed 3/10/2026).
Claims 1 and 7
Irving teaches a deep neural network-based disease-of-interest prediction device, the device comprising: at least one processor; and a memory storing instructions that, when executed by the at least one processor, cause the at least one processor to (Irving abstract, Fig. 9, [0070]-[0079], noting a processor device executing various program modules to train and use a neural network-based architecture to predict patients with a pathology of interest, e.g. heart failure as in [0011]):
collect medical diagnosis data of a plurality of patients (Irving [0031], noting the system receives an input data structure corresponding to electronic health record data including physiological measurements, medical codes, procedural codes, etc. (i.e. medical diagnosis data) of one or more patient cohorts);
generate input data by embedding the medical diagnosis data of the plurality of patients into a (Irving [0032], noting the input data structure is pre-processed to generate a vector of n x m dimensions, with n being the number of patients in the cohort and m being the number of EHR features for each patient);
generate a deep neural network-based disease-of-interest prediction model comprising an autoencoder having a bottleneck layer and a classification layer connected to the bottleneck layer (Irving [0005], [0033], [0037]-[0038], [0046], noting the vector is processed to train (i.e. generate) an autoencoder neural network architecture by learning a lower-dimensional embedding (i.e. bottleneck) layer which feeds into a classification layer to predict the pathology);
train the deep neural network-based disease-of-interest prediction model by:
(i) inputting the input data into the autoencoder to generate compressed data at the bottleneck layer and reconstructing the input data based on the compressed data (Irving [0033], [0037]-[0038], noting the autoencoder is trained to generate low-dimensional (i.e. compressed) embeddings (i.e. at a bottleneck layer) from the input data and reconstruct the input data from the low-dimensional embeddings);
(ii) predicting a development of disease of interest via the classification layer based on the compressed data (Irving [0005], [0046], noting classification layer uses the low-dimensional embeddings to predict the pathology of interest); and
(iii) optimizing the autoencoder and the classification layer simultaneously to minimize a final cost function, wherein the final cost function is a linear sum of a first cost function calculating a reconstruction error of the autoencoder and a second cost function calculating a prediction error of the classification layer, the linear sum applying individual weights to the first cost function and the second cost function (Irving [0005], [0010], [0056]-[0057], noting the system calculates and optionally combines reconstruction error of the autoencoder and classification error of the classification layer, e.g. via a linear weighted sum equation that applies individual weights to each of the reconstruction loss and the classification loss to calculate an overall loss L which is then used for optimizing the training of the neural network architecture layers, e.g. by minimizing the calculated loss function L as noted in [0040]; such an optimization process is considered to simultaneously optimize the autoencoder and classification layers because errors from both of these layers are being considered when performing updated training (i.e. optimization) across the entire model as in [0005] & [0010]),
wherein a future development of disease of interest of a patient is predicted using the generated deep neural-network disease-of-interest prediction model based on existing medical diagnosis data of the patient (Irving [0012], noting the trained neural network architecture can be used to classify input diagnostic data, e.g. by predicting a pathology of interest like heart failure as in [0011]).
In summary, Irving teaches a system for training and using an autoencoder neural network architecture to process input data including EHR data and output a predicted pathology of interest. The input EHR data is pre-processed into a vector representing m dimensions of EHR data for each patient as disclosed in [0032]; however, Irving fails to explicitly disclose that the m-dimensional vector is a binary vector. However, Vasiljeva teaches that EHR data including ICD codes can be processed into a binary vector of fixed length reflecting the presence or absence of a given code in the patient’s EHR for the purpose of disease prediction modeling (Vasiljeva sections 1.1 & 2.3). It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the EHR vectorization process of Irving such that it results in a vector with binary feature representations as in Vasiljeva in order to reduce the size of the space over which prediction takes place while still maintaining the predictive power of presence of a given feature in a patient’s history (as suggested by Vasiljeva section 2.3).
Claim 7 recites substantially similar subject matter as claim 1, and is also rejected as above.
Claims 2 and 8
Irving in view of Vasiljeva teaches the device of claim 1, and the combination further teaches wherein the at least one processor extracts disease code information assigned to each patient from the medical diagnosis data, and wherein the disease code information is International Statistical Classification of Disease (ICD) codes assigned to the each patient (Irving [0031], noting the EHR data includes medical codes, e.g. ICD codes as shown in Fig. 7 & [0069]; see also Vasiljeva section 1.1).
Claim 8 recites substantially similar subject matter as claim 2, and is also rejected as above.
Claims 3 and 9
Irving in view of Vasiljeva teaches the device of claim 2, and the combination further teaches wherein the at least one processor counts types of the ICD codes assigned to the each patient, and generates the input data for the each patient as a binary vector having a size corresponding to types of the ICD codes, and wherein the input data determines a binary value of the binary vector based on existence of a diagnosis history for each of the ICD codes for the each patient (Irving [0032], noting generation of an m-dimensional vector for each patient, where m corresponds to the number of EHR data points (which can include ICD codes per Fig. 7 and when considered in the context of the combination with Vasiljeva) observed over the patient cohort, indicating that the system counts these different types of features (e.g. ICD codes) to determine the value of m; see also Vasiljeva section 2.3, noting binary vectorization of ICD codes such that the presence or absence of a given code in the patient’s EHR is embedded into the fixed length vector (analogous to the m-dimensional vector of Irving)).
Claim 9 recites substantially similar subject matter as claim 3, and is also rejected as above.
Claim 5
Irving in view of Vasiljeva teaches the device of claim 1, and the combination further teaches wherein the autoencoder comprises an encoder that maps the input data into a latent space dimension to output the compressed data toward the bottleneck layer, and a decoder configured to reconstruct the compressed data of the bottleneck layer into the input data (Irving [0033], [0037]-[0038], noting the autoencoder includes an encoder to dimensionally reduce (i.e. compress) the data into an embeddings layer with the lowest number of dimensions (i.e. a bottleneck layer) and a decoder that reconstructs the input data from the low-dimensional embeddings), and wherein the classification layer is configured with a multi-layer perceptron structure connected to the bottleneck layer to predict whether a disease of interest has developed through supervised learning that inputs the compressed data of the bottleneck layer and outputs correct answer data corresponding to the compressed data (Irving [0005], [0037], [0046]-[0047], noting the neural network architecture includes a feedforward multilayer perceptron structure including a fully-connected classification layer trained via supervised learning to use the low-dimensional embeddings to predict the pathology of interest).
Claim 11
Irving in view of Vasiljeva teaches a non-transitory computer-readable recording medium storing a set of instruction (Irving [0075]) that, when executed by one or more processors of a deep neural network-based disease-of-interest prediction device, cause the deep neural network-based disease-of-interest prediction device to perform the deep neural network-based disease-of-interest prediction method of claim 7 (see explanation for claims 1 and 7 above).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Miotto et al. (US 20200327404 A1) describes systems and methods for training and using autoencoder neural network diagnosis models to evaluate binarized vectors of diagnosis code data.
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 KAREN A HRANEK whose telephone number is (571)272-1679. The examiner can normally be reached M-F 8:00-4:00 ET.
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, Shahid Merchant can be reached at 571-270-1360. 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.
/KAREN A HRANEK/ Primary Examiner, Art Unit 3684
Prosecution Insights