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 .
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) 4-13, 18 are rejected under 35 U.S.C. 103 as being unpatentable McNair (11,694,814) in view of He et al (20240362286).
As per claim 1, McNair (11,694,814) teaches a method for enhancing natural language processing (NLP) model training in biomedical context (as, using in EHR systems – col. 2 lines 14-22 , to improve patient information/services including natural language processing -- col. 3 lines 55-62), the method comprising:
training an embedding model to capture semantic richness of biomedical context based, at least in part, on a set of unstructured texts obtained from an electronic health record (EHR) system (as, using unstructured text toward EHR –col. 2 lines 19-31);
obtaining a seed set including seed texts each associated with a classification label; obtaining an unlabeled set including unlabeled texts;
for each seed text of the seed set, inputting at least a portion of the seed text into the trained embedding model to generate a respective vectorized semantic representations of the seed text(as, using historical patient records to develop a structural topic modeling pertaining to subjects such as observational finding, complaints, clinical assessments, program, diagnostics, etc – col. 3 line13-18, and developing reference clusters – col. 3 lines 20-25; the ‘seed set’ are either the topics modeled, or the developed clusters);
for each unlabeled text of the unlabeled set, in putting at least a portion of the unlabeled text into the trained embedding model to generate a respective vectorized semantic representation of the unlabeled text (as performing a vectorized cosine of elements measuring the distance between the historical texts and the current/unstructured text – col. 3 lines 26-40);
assigning classification labels to at least a subset of the unlabeled set by clustering the seed set with the unlabeled set based (col. 6 lines 22-40 – clustering of the unstructured text according to similar topics),
at least in part, on the respective vectorized semantic representations for each seed text and for each unlabeled text (as performing a vectorized cosine of elements measuring the distance between the historical texts and the current/unstructured text – col. 3 lines 30-40); and providing training data including the assigned classification labels, for training an NLP model that is different than the embedding model (and training the model based on the above calculations – fig. 3a, using unstructured patient text to train the system to match diagnoses’/treatment, see also col. 5 lines 60 – col. 6 line 9 – wherein the unstructured narrative text is used to train the structure topic model to discover latent topics).
As per the relationship between the seed text and the unlabeled text (esp. towards size and the implication that the ‘set of seed text’ is smaller in size), McNair (11,694,814) teaches the concept of using historical data to develop a set of clusters/groupings, then using the modeling to recognize unstructured text; although intimated, McNair (11,694,814) does not explicitly teach the use of a seed set that is smaller in size (the technique found in McNair (11,694,814) is a version of “semi-supervised”, but does not explicitly state as such); He et al (20240362286) teaches the use of semi-supervised ML modeling, wherein a small amount of labeled known data, which improves the accuracy and with unstructured/unlabeled data, achieving higher generalization. Therefore, it would have been obvious to one of ordinary skill in the art of ML modeling to further specify the training/learning technique as disclosed in McNair (11,694,814) (especially the referred-to-section of column 3) to use a smaller amount of structured/known data with a larger amount of unstructured/unlabeled data, as taught by He et al (20240362286), because it would advantageously provide higher accuracy, better generalization, while managing the cost of the model data ( He et al (20240362286) , para 0084).
As per claim 2, the combination of McNair (11,694,814) in view of He et al (20240362286) teaches the method of claim 1, wherein the set of unstructured texts includes context snippets identified from clinical notes (see McNair (11,694,814), as taking sections/chunks – fig. 3a, of clinical narrative data – col. 2, lines 27-31).
As per claim 3, the combination of McNair (11,694,814) in view of He et al (20240362286) teaches the method of claim 2, wherein the context snippets are identified based, at least in part, on at least one of demographics, medical history, diagnosis, severity of disease, medication, therapy, surgery, or associated outcome (see McNair (11,694,814), as, the narrative data leads to clinical diagnoses – col. 2 lines 33-49).
As per claim 14, the combination of McNair (11,694,814) in view of He et al (20240362286) teaches the method of claim 1, wherein assigning classification labels to at least a subset of the unlabeled set by clustering the seed set with the unlabeled set comprises iteratively performing the clustering while expanding the seed set (see McNair (11,694,814), as starting with known cluster membership and then expanding condition-associated cluster determined by the STM modeling – col. 3 lines 19-31).
As per claim 15, the combination of McNair (11,694,814) in view of He et al (20240362286) teaches the method of claim 1, wherein the classification labels assigned to the at least a subset of the unlabeled set is based, at least in part, on one or more nearest neighbors in a finalized seed set (see McNair (11,694,814), as using the labeling – col. 3 lines 33-36; using distance/nearest neighbor measurement – col. 3 lines 23-29).
As per claim 16, the combination of McNair (11,694,814) in view of He et al (20240362286) teaches the method of claim 1, wherein at least a subset of the assigned classification labels is used to generate, modify, or supplement structured texts in the EHR system (see McNair (11,694,814), col. 3 lines 8-18, operating on unstructured medical records using structure topic modeling).
As per claim 17, the combination of McNair (11,694,814) in view of He et al (20240362286) teaches the method of claim 1, wherein at least a subset of the assigned classification labels is used in conjunction with data obtained from EHR system (see McNair (11,694,814), as applied to HER systems – col. 9 lines 3-15) as input into at least one of a classification, prediction, or association model to produce output (as labeled clusters of the data – col. 3 lines 30-39) and applied to the incoming unstructured data using structure topic modeling – col. 3 lines 10-15).
Claims 19,20 are system/non-transitory computer readable medium claims that perform the steps found throughout claims 1-3, 14-17 above and as such, claims 19,20 are similar in scope and content to claims 1-3, 14-17; therefore claims 19,20 are rejected under similar rationale as presented against claims 1-3,14-17 above. Furthermore, McNair (11,694,814) teaches systems with processors and memories executing the referenced steps in McNair – see Fig. 1a, subblock 120,121, and 175 (cpu device, storage device, and a network, respectively).
Claim(s) 4-13, 18 are rejected under 35 U.S.C. 103 as being unpatentable over the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320).
Regarding the McNair (11,694,814) reference, McNair (11,694,814) teaches the general concepts of using distributed network and processing packages – for e.g., col. 12 lines 35-43, discussing differing server structures for databases and ontologies, see also col. 12 lines 60-67; see also col. 13 lines 5-27; further, McNair (11,694,814) teaches mixed membership models utilizing global prior distribution with a derived topic from a current word, and then developing posterior distribution that can be re-initialized for different solutions (col. 12 lines 31-42, lines 45-48; other techniques, such as STM, allows the analyst to incorporate covariates during the E-M steps (col. 12 lines 57-64); and as mapped to claim 1, McNair (11,694,814) teaches using a structured topic model in analyzing topics within the unstructured data (col. 3 lines 12-30).
As per claims 4,5, 18, the combination of McNair (11,694,814) in view of He et al (20240362286) does not explicitly teach a labeled “Large Language Model” with using tokenized masks operating on the unstructured data, however, Renne et al (20250174320) teaches operating/processing narrated medical reports (para 0016), and hence unstructured text, operated on by masking tokens as part of a pre-trained transform network in a Large Language Model (Renne et al (20250174320) , para 0092, and para 0095, 0097). Therefore, it would have been obvious to one of ordinary skill in the art of medical record/transcript processing to modify the unstructured text processing of McNair (11,694,814) in view of He et al (20240362286) with a type of pretrained transformer network using tokenized masks in a LLM structure, as taught by Renne et al (20250174320) , because it would advantageously improve training time and accuracy of the layers (see Renne et al (20250174320), para 0095) as well as masking personal information embedded in the reports (see Renne et al (20250174320), para 0017).
Further to claim 4, the combination of McNair (11,694,814) in view of He et al (20240362286) in view of Renne et al (20250174320) teaches fine-tuning the models (see McNair (11,694,814) col. 16 lines 7-12; and Renne et al (20250174320) – para 0099).
As per claim 5, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 4, further comprising, for each unstructured text of the set of unstructured texts: extracting one or more entities of interest from the unstructured text; and replacing at least a subset of the one or more extracted entities of interest with one or more types of mask tokens in the unstructured text to generate respective masked text (see combination statement above, as well as, (Renne et al (20250174320) , para 0092, and para 0095, 0097 teaching the mask tokens to replace/hide the personal information being masked).
As per claim 6, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 5, wherein the at least a subset of the one or more extracted entities is selected based, at least in part, on a task of an NLP model to be trained with the NLP model training data (as operating on, and training on, using natural language phrases – see para 0016, 0019, in Renne et al (20250174320)).
As per claim 7, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 6, wherein the task of the NLP model includes at least one of predicting diagnoses, identifying biomarker, recommending treatment, or determining medication intake (as analyzing medical records, including treatment plans -- Renne et al (20250174320) , para 0019).
As per claim 8, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 5, further comprising labeling each masked text with at least one target label based, at least in part, on the one or more entities of interest extracted from the unstructured text (see Renne et al (20250174320), as mapped above in the combination, for the details of the masking, and further, in para 0022, wherein the masks removes personal information; the ‘target label’ is the topic of “personalized information”).
As per claim 9, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 8, wherein the labeling comprises normalizing the one or more entities of interest based, at least on part, on medical or clinical ontology (as Renne et al (20250174320) operating on medical data in a repository – para 0002, including a normalization based on tokens tied to the embedded topics of personalized information – see para 0095 for normalization, see para 0094 for the topic specific limitations).
As per claim 10, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 8, further comprising adapting the LLM to use each masked text to predict its associated target label (see Renne et al (20250174320), using LLM formats – para 0092).
As per claim 11, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 10, wherein the parameters of the LLM are adjusted during the adapting and fixed after the adapting is completed (see Renne et al (20250174320) , para 0092, operating on LLM’s, wherein after the fine-tuning – para 0099, para 0103-0106, remaining fixed after the supervised fine-tuning).
As per claim 12, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 1, wherein given an input to the trained embedding model, a vectorized semantic representation of the input is generated based, at least in part, on output of one or more layers of the trained embedding model (as performing semantic coherence on the topic fine tuning – see McNair (11,694,814), col. 15 line 55 – col. 16 line 7).
As per claim 13, the combination of McNair (11,694,814) in view of He et al (20240362286), in further view of Renne et al (20250174320) teaches the method of claim 12, wherein the vectorized semantic representation of the input is generated by averaging the output of the one or more layers (see McNair (11,694,814) teaching altering the variance/averaging by the analyst – col. 14 lines 55-64, using global variables – which includes the above semantic coherence; the application of McNair (11,694,814) in a layered transformer network, is taught and mapped above, in the Renne et al (20250174320) reference).
Further to claim 18, the combination of McNair (11,694,814) in view of Renne et al (20250174320) as applied to claims 4,5 above, shows the LLM (the listing of elements in claim 18 are in the alternative, and hence, support for teaching the claimed LLM meets the claim scope of claim 18.)
Response to Arguments
Applicant’s arguments with respect to the claim(s) have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument. Examiner notes the introduction of the He et al (20240362286) reference to meet the newly amended claim language, more toward the relationship between seed text and unlabeled text, as well as, further explanation/mapping to the McNair (11,694,814) reference.
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.
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Please see related art listed on the PTO-892 form.
Furthermore, the following references were found that teach various features found in the claims/specification:
For the concept of semi-supervised ML for medical records/field --
See Soori-arachi (20250006376), para 0330, for the AI analysis algorithms
See Paik et al (20240177836) teaching semi-supervised learning of EHR (para 0255, processing data from HER – para 0322)
See Gupta et al (20240029848) teaching semi-supervised ML generating text reports on health care plans (para 0226, 0235)
For other aspects of ML in the field of EHR:
Shah et al (20250201393) teaches the use of support vector machines in a LLM environment to process EHR unstructured text (para 0053 – 0055)
Kohl et al (20250173508) teaches neural network models processing EHR using masking tokens in a LLM transformer structure (para 0057, 0063, 0119).
Ghosh et al (20250149187) teaches NLP of EHR records (para 0027-0028) using masking tokens (para 0039).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Michael Opsasnick, telephone number (571)272-7623, who is available Monday-Friday, 9am-5pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Mr. Richemond Dorvil, can be reached at (571)272-7602. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/Michael N Opsasnick/Primary Examiner, Art Unit 2658 05/26/2026
Prosecution Insights