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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 10/27/2023, is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
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 1-20 are rejected under 35 U.S.C. § 101 because the claimed invention is directed to a judicial exception (i.e. an abstract idea) without significantly more.
Step 1 – Statutory Categories of Invention:
Claims 1, 8, and 15 are drawn to method, a system, and an article of manufacture, which are statutory categories of invention.
Step 2A – Judicial Exception Analysis, Prong 1:
Independent claims 1 recites a method comprising obtaining an original dataset to generate a synthetic doctor-patient conversation, the original dataset including data in the form of textual dialogue associated with each of a plurality of individual doctor-patient conversations; constructing input data by performing on the dialogue of the original dataset that captures and categorizes named medical entities present in the dialogue; generating prepared input data by arranging the input data in an annotated turn-by-turn conversation format; randomly selecting a disease from a medical knowledge graph and capturing a plurality of symptoms that are mapped to the selected disease; and to generate a turn-by-turn synthetic doctor- patient conversation, by inputting the plurality of symptoms as a first control parameter of the and randomly assigning a first speaker.
Independent claim 8 recites a system comprising obtaining an original dataset to generate a synthetic doctor-patient conversation, the original dataset including data in the form of textual dialogue associated with each of a plurality of individual doctor-patient conversations; constructing input data by performing on the dialogue of the original dataset that captures and categorizes named medical entities present in the dialogue; generating prepared input data by arranging the input data in an annotated turn-by-turn conversation format; randomly selecting a disease from a medical knowledge graph and capturing a plurality of symptoms that are mapped to the selected disease; and to generate a turn-by-turn synthetic doctor- patient conversation, by inputting the plurality of symptoms as a first control parameter of the and randomly assigning a first speaker.
Independent claim 15 recites an article of manufacture comprising obtaining an original dataset to generate a synthetic doctor-patient conversation, the original dataset including data in the form of textual dialogue associated with each of a plurality of individual doctor-patient conversations; constructing input data by performing on the dialogue of the original dataset that captures and categorizes named medical entities present in the dialogue; generating prepared input data by arranging the input data in an annotated turn-by-turn conversation format; randomly selecting a disease from a medical knowledge graph and capturing a plurality of symptoms that are mapped to the selected disease; and to generate a turn-by-turn synthetic doctor- patient conversation, by inputting the plurality of symptoms as a first control parameter of the and randomly assigning a first speaker.
These steps amount to certain methods of organizing human activity which includes functions relating to managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions) (MPEP § 2106.04(a)(2)(II)(C) citing the abstract idea grouping for methods of organizing human activity for managing personal behavior or relationships or interactions between people – also note MPEP § 2106.04(a)(2)(II) stating certain activity between a person and a computer may fall within the “certain methods of organizing human activity” grouping).
Step 2A – Judicial Exception Analysis, Prong 2:
This judicial exception is not integrated into a practical application because the additional elements within the claims only amount to instructions to implement the judicial exception using a computer [MPEP 2106.05(f)].
Claim 1 recites computer-implemented. Claims 8 and 16 recite one or more data processors and one or more non-transitory computer readable storage media.
These elements are recited at a high-level of generality such that it amounts to mere instructions to apply the exception because this is an example of applying the
abstract idea by use of general-purpose computer which does not integrate the abstract
idea into a practical application.
Claims 1, 8, and 15 recite for training a machine learning model, a named entity recognition operation, using an input data preparation algorithm having various control parameters; training the machine learning model using the prepared input data; a conversation generation control algorithm having various control parameters, machine learning model, and conversation generation control algorithm as tools to apply data to an algorithm and report the results (MPEP § 2106.05(f)(2) see case involving a commonplace business method or mathematical algorithm being applied on a general-purpose computer within the “Other examples.. i.”) amounting to instruction to implement the abstract idea using a general-purpose computer. Alice Corp. Pty. Ltd. V. CLS Bank Int’l, 134 S. Ct. 2347, 1357 (2014).
The above claims, as a whole, are therefore directed to an abstract idea.
Step 2B – Additional Elements that Amount to Significantly More:
The present claims do not include additional elements that are sufficient to amount to more than the abstract idea because the additional elements or combination of elements amount to no more than a recitation of instructions to implement the abstract idea on a computer.
Claim 1 recites computer-implemented. Claims 8 and 16 recite one or more data processors and one or more non-transitory computer readable storage media.
Thus, taken alone, the additional elements do not amount to significantly more than the above-identified judicial exception. Looking at the limitations as an ordered combination adds nothing that is not already present when looking at the elements taken individually. Their collective functions merely provide conventional computer implementation.
Claims 1, 8, and 15 recite for training a machine learning model, a named entity recognition operation, using an input data preparation algorithm having various control parameters; training the machine learning model using the prepared input data; a conversation generation control algorithm having various control parameters, machine learning model, and conversation generation control algorithm as tools to apply data to an algorithm and report the results (MPEP § 2106.05(f)(2) see case involving a commonplace business method or mathematical algorithm being applied on a general-purpose computer within the “Other examples.. i.”) amounting to instruction to implement the abstract idea using a general-purpose computer. Alice Corp. Pty. Ltd. V. CLS Bank Int’l, 134 S. Ct. 2347, 1357 (2014).
For the reasons stated, these claims are consequently rejected under 35 U.S.C. § 101.
Analysis of Dependent Claims
Dependent claim(s) 2, 9, and 16 recite(s) wherein the named medical entities are selected from the group consisting of biomedical information, personal identifying information, personal health information, and combinations thereof.
Dependent claim(s) 5 recite(s) wherein generating the prepared input data further comprises: comparing the total token length of the context parameter value to a predetermined maximum allowable context token length value; and when the total token length of the context parameter value exceeds the predetermined maximum allowable context token length value, remove dialogue turns one at a time from the context parameter until the total token length value of the context parameter no longer exceeds the maximum allowable context token length value.
Dependent claim(s) 11 and 18 recite(s) wherein the format of the prepared input data is: [TOPICS] <symptom values> [ENTITIES] <entity values> [SPEAKER] <doctor/patient> [TURN LENGTH] <length of the turn>[REMAINING_TURNS] <number of turns left> [CONTEXT] <previous dialogues if available>.
Each of these steps of the preceding dependent claims 2, 5, 9, 11, 16, and 18 only serve to further limit or specify the features of independent claims 1, 8, and 15 accordingly, and hence are nonetheless directed towards fundamentally the same abstract idea as the independent claim and utilize the additional elements analyzed below in the expected manner.
Dependent claim(s) 3 recite(s) wherein constructing the input data further comprises: post-processing the named medical entities to remove any tagging anomalies resulting from the named entity recognition operation; and performing an additional medical named entity recognition operation on the post- processed named medical entities using at least one model trained on a biomedical corpus. The named entity recognition operation and performing an additional medical named entity recognition operation on the post- processed named medical entities using at least one model trained on a biomedical corpus are recited as tools to apply data to an algorithm and report the results (MPEP § 2106.05(f)(2) see case involving a commonplace business method or mathematical algorithm being applied on a general-purpose computer within the “Other examples.. i.”) amounting to instruction to implement the abstract idea using a general-purpose computer. Alice Corp. Pty. Ltd. V. CLS Bank Int’l, 134 S. Ct. 2347, 1357 (2014).
Dependent claim(s) 4, 10, and 17 recite(s) wherein: the control parameters of the input data preparation algorithm and the conversation generation control algorithm include symptoms/topics, entities sampled from the symptoms/topics, speaker identity, conversation turn length, remaining conversation turns, and context; and wherein a value of the context control parameter for a given conversation turn comprises the cumulative dialogue of all the preceding conversation turns, when there is at least one preceding conversation turn. The input data preparation algorithm and conversation generation control algorithm are recited as tools to apply data to an algorithm and report the results (MPEP § 2106.05(f)(2) see case involving a commonplace business method or mathematical algorithm being applied on a general-purpose computer within the “Other examples.. i.”) amounting to instruction to implement the abstract idea using a general-purpose computer. Alice Corp. Pty. Ltd. V. CLS Bank Int’l, 134 S. Ct. 2347, 1357 (2014). wherein a value of the context control parameter for a given conversation turn comprises the cumulative dialogue of all the preceding conversation turns, when there is at least one preceding conversation turn is part of the abstract idea, which further specifies and limits the independent claims.
Dependent claim(s) 6, 12, and 19 recite(s) wherein training the machine learning model includes multistage finetuning of the machine learning model, the multistage finetuning comprising: first-fold finetuning using a medical question answering dataset; following first-fold finetuning, second-fold finetuning guided by the input data preparation algorithm and the control parameters thereof and using a dataset including data in the form of textual dialogue associated with each of a plurality of individual non-medical specific conversations; and following second-fold finetuning, final-fold finetuning guided by the input data preparation algorithm and the control parameters thereof and using a dataset including data in the form of textual dialogue associated with each of a plurality of individual medical-specific conversations, wherein the symptoms/topics control parameter of the input data preparation algorithm is composed of keywords previously derived from a dataset including data in the form of textual dialogue associated with each of a plurality of individual non-medical conversations. This limitation is recited as tools to apply data to an algorithm and report the results (MPEP § 2106.05(f)(2) see case involving a commonplace business method or mathematical algorithm being applied on a general-purpose computer within the “Other examples.. i.”) amounting to instruction to implement the abstract idea using a general-purpose computer. Alice Corp. Pty. Ltd. V. CLS Bank Int’l, 134 S. Ct. 2347, 1357 (2014).
Dependent claim(s) 7 and 13 recite(s) comprising evaluating performance of the trained machine learning model by analyzing the generated turn-by-turn synthetic doctor-patient conversation using a metric selected from the group consisting of recall- oriented understudy for gisting evaluation, n-gram diversity score, unique n-gram count, and combinations thereof. The trained machine learning model is recited as a tool to apply data to an algorithm and report the results (MPEP § 2106.05(f)(2) see case involving a commonplace business method or mathematical algorithm being applied on a general-purpose computer within the “Other examples.. i.”) amounting to instruction to implement the abstract idea using a general-purpose computer. Alice Corp. Pty. Ltd. V. CLS Bank Int’l, 134 S. Ct. 2347, 1357 (2014).
Dependent claim(s) 14 and 20 recite(s) wherein the machine learning model has a transformer- encoder-decoder architecture. This limitation is recited as a tool to apply data to an algorithm and report the results (MPEP § 2106.05(f)(2) see case involving a commonplace business method or mathematical algorithm being applied on a general-purpose computer within the “Other examples.. i.”) amounting to instruction to implement the abstract idea using a general-purpose computer. Alice Corp. Pty. Ltd. V. CLS Bank Int’l, 134 S. Ct. 2347, 1357 (2014).
Claim Rejections - 35 USC § 103
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) 1-5, 8-11, 14-18, and 20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Tang (Terminology-aware Medical Dialogue Generation) in view of Wang (US 20210082585 A1) in view of Varshney (Knowledge graph assisted end-to-end medical dialog generation).
REGARDING CLAIM 1
Tang teaches a computer-implemented method comprising:
obtaining an original dataset for training a machine learning model to generate a synthetic doctor-patient conversation, the original dataset including data in the form of textual dialogue associated with each of a plurality of individual doctor-patient conversations; ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.)
generating prepared input data by arranging the input data in an annotated turn-by-turn conversation format using an input data preparation algorithm having various control parameters; ([Pg. 2 2.2. Terminological Knowledge Enhancement] Obtain a corpus containing nearly 250,000 doctor-patient conversation pairs with 3,600M tokens. We set up our experiment based on this terminology-enhanced dataset, whose statistics
are shown in Table 1. Table 1 showcases data statistics of the Medical English Dialogue Corpus with annotated medical terminologies (abbr. Terms).)
training the machine learning model using the prepared input data; ([Pg. 2 2.3 Response Generation] We employ BART (encoder-decoder structure) as the base model. [Pg. 3 2.4 and 2.5] An auxiliary classification task is introduced to train the encoder to identify terminology-related tokens. Both aforementioned task objectives are combined to train the language model, and the overall loss function Loverall is minimized during model fine-tuning:)
and causing the trained machine learning model to generate a turn-by-turn synthetic doctor- patient conversation according to a conversation generation control algorithm having various control parameters, by inputting the plurality of symptoms to the machine learning model as a first control parameter of the conversation generation control algorithm and randomly assigning a first speaker. ([Pg. 2 2.1 Task Definition] The given inputs are in the form of a text sequence X = {x1, x2, ..., xn}, which consists of a patient’s question alongside the collection of prior dialogue turns between the doctor and the patient. The goal of the task is to generate a response Y = {y1, y2, ..., ym} (as a doctor) by modeling the conditional probability distribution P(Y |X). ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.))
Tang does not explicitly teach, however Wang teaches
constructing input data by performing a named entity recognition operation on the dialogue of the original dataset that captures and categorizes named medical entities present in the dialogue; ([Para. 0026] The output of section extractor 231, sections of conversation 240, is the input to named entity recognition engine 241, and thus, for illustrative purposes, section extractor 231 is shown as overlapping with, or intersecting, named entity recognition engine 241, precisely at sections of conversation 240. Named entity recognition engine 241 is configured to recognize various types of words or phrases of relevance to generating a medical record according to the hard medical record format, such as, for example, a symptom, a diagnosis, a drug, or an occupation, as shown in FIG. 4 at block 450, for example. The output of named entity recognition engine 241 is medical entities 250.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang and incorporate generation of medical records based on doctor-patient dialogue as taught by Wang, with the motivation of facilitating patient care (Wang Para. 0002).
Tang/ Wang does not explicitly teach, however Varshney teaches
randomly selecting a disease from a medical knowledge graph and capturing a plurality of symptoms that are mapped to the selected disease; ([Pg. 1 Abstract] The medical-specific knowledge graph contains broadly 3 types of medical-related information, including disease, symptom and laboratory test.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang, generation of medical records based on doctor-patient dialogue as taught by Wang, and incorporate a medical knowledge graph as taught by Varshney, with the motivation of improving access to healthcare services, improving patient treatment quality, and lowering medical expenses (Varshney Abstract).
REGARDING CLAIM 2
Tang/ Wang/ Varshney teach the computer-implemented method of claim 1, Wang further teaches wherein the named medical entities are selected from the group consisting of biomedical information, personal identifying information, personal health information, and combinations thereof. ([Para. 0036] The medical entities relevant to this embodiment are, symptom, diagnosis, drug and occupation.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang and incorporate generation of medical records based on doctor-patient dialogue as taught by Wang, with the motivation of facilitating patient care (Wang Para. 0002).
REGARDING CLAIM 3
Tang/ Wang/ Varshney teach the computer-implemented method of claim 1, Wang further teaches wherein constructing the input data further comprises: post-processing the named medical entities to remove any tagging anomalies resulting from the named entity recognition operation; and performing an additional medical named entity recognition operation on the post- processed named medical entities using at least one model trained on a biomedical corpus. ([Para. 0035] It is noted with reference to FIGS. 2 and 3 that medical entity 330 of FIG. 3 is the same as named entities 250, generated by named entity recognition engine 241, both of FIG. 2. Thus, in the process 300 of FIG. 3, medical entities 330, generated as shown in FIG. 2, may be used to assign a QA pair, or, for example, an isolated sentence, to a section type 350. Based on the section type 350, different medical entities are placed into the correct fields of a hard medical record template. It is noted that for the hard format medical record, a key task is filling the hard record template with medical entities. FIG. 4 illustrates the example main portion 120 of example dialogue 100 of FIG. 1 after being divided into three sections, with named medical entities identified in each section, according to one embodiment disclosed herein. The divided example dialogue of FIG. 4 is the result of processing each of the 4 QA pairs of main portion 120, and the isolated doctor's final summation statement of main portion 120, using elements 210 through 250 of system 200 of FIG. 2, which includes the use of process 300 of FIG. 3 by section extractor 231.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang and incorporate generation of medical records based on doctor-patient dialogue as taught by Wang, with the motivation of facilitating patient care (Wang Para. 0002).
REGARDING CLAIM 4
Tang/ Wang/ Varshney teach the computer-implemented method of claim 1, Tang further teaches wherein:
the control parameters of the input data preparation algorithm and the conversation generation control algorithm include symptoms/topics, entities sampled from the symptoms/topics, speaker identity, conversation turn length, remaining conversation turns, and context; ([Pg. 2 2.1 Task Definition] The given inputs are in the form of a text sequence X = {x1, x2, ..., xn}, which consists of a patient’s question alongside the collection of prior dialogue turns between the doctor and the patient. ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.))
and wherein a value of the context control parameter for a given conversation turn comprises the cumulative dialogue of all the preceding conversation turns, when there is at least one preceding conversation turn. ([Pg. 2 2.2. Terminological Knowledge Enhancement] Obtain a corpus containing nearly 250,000 doctor-patient conversation pairs with 3,600M tokens. We set up our experiment based on this terminology-enhanced dataset, whose statistics are shown in Table 1. Table 1 showcases data statistics of the Medical English Dialogue Corpus with annotated medical terminologies (abbr. Terms).)
REGARDING CLAIM 5
Tang/ Wang/ Varshney teach the computer-implemented method of claim 4, Varshney further teaches wherein generating the prepared input data further comprises: comparing the total token length of the context parameter value to a predetermined maximum allowable context token length value; ([Pg. 3 2. Related Work] Used BART as their base model, and to predict the responses, they encoded the characteristics of each input sequence using self-attention and then decoded them autoregressively. During training, the predicted tokens were constrained to match the gold standard responses provided by actual doctors. This allowed the model to learn from expert feedback and improve its output.)
and when the total token length of the context parameter value exceeds the predetermined maximum allowable context token length value, remove dialogue turns one at a time from the context parameter until the total token length value of the context parameter no longer exceeds the maximum allowable context token length value. ([pg. 7, 5.4 Error Analysis] The reason being the truncation of input and output sequences since maximum sequence size is restricted to a fixed length for the pre-trained models.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang, generation of medical records based on doctor-patient dialogue as taught by Wang, and incorporate a medical knowledge graph as taught by Varshney, with the motivation of improving access to healthcare services, improving patient treatment quality, and lowering medical expenses (Varshney Abstract).
REGARDING CLAIM 8
Tang a system comprising:
obtaining an original dataset for training a machine learning model to generate a synthetic doctor-patient conversation, the original dataset including data in the form of textual dialogue associated with each of a plurality of individual doctor-patient conversations; ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.)
generating prepared input data by arranging the input data in an annotated turn- by-turn conversation format using an input data preparation algorithm having various control parameters; ([Pg. 2 2.2. Terminological Knowledge Enhancement] Obtain a corpus containing nearly 250,000 doctor-patient conversation pairs with 3,600M tokens. We set up our experiment based on this terminology-enhanced dataset, whose statistics
are shown in Table 1. Table 1 showcases data statistics of the Medical English Dialogue Corpus with annotated medical terminologies (abbr. Terms).)
training the machine learning model using the prepared input data; ([Pg. 2 2.3 Response Generation] We employ BART (encoder-decoder structure) as the base model. [Pg. 3 2.4 and 2.5] An auxiliary classification task is introduced to train the encoder to identify terminology-related tokens. Both aforementioned task objectives are combined to train the language model, and the overall loss function Loverall is minimized during model fine-tuning:)
and causing the trained machine learning model to generate a turn-by-turn synthetic doctor-patient conversation according to a conversation generation control algorithm having various control parameters, by inputting the plurality of symptoms to the machine learning model as a first control parameter of the conversation generation control algorithm and randomly assigning a first speaker. ([Pg. 2 2.1 Task Definition] The given inputs are in the form of a text sequence X = {x1, x2, ..., xn}, which consists of a patient’s question alongside the collection of prior dialogue turns between the doctor and the patient. The goal of the task is to generate a response Y = {y1, y2, ..., ym} (as a doctor) by modeling the conditional probability distribution P(Y |X). ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.))
Tang does not explicitly teach, however Wang teaches
one or more data processors; ([Para. 0006] one or more computer processors)
and one or more non-transitory computer readable media storing instructions which, when executed by the one or more data processors, cause the one or more data processors to perform the following operations: ([Para. 0006] a computer-readable storage medium)
constructing input data by performing a named entity recognition operation on the dialogue of the original dataset that captures and categorizes named medical entities present in the dialogue; ([Para. 0026] The output of section extractor 231, sections of conversation 240, is the input to named entity recognition engine 241, and thus, for illustrative purposes, section extractor 231 is shown as overlapping with, or intersecting, named entity recognition engine 241, precisely at sections of conversation 240. Named entity recognition engine 241 is configured to recognize various types of words or phrases of relevance to generating a medical record according to the hard medical record format, such as, for example, a symptom, a diagnosis, a drug, or an occupation, as shown in FIG. 4 at block 450, for example. The output of named entity recognition engine 241 is medical entities 250.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang and incorporate generation of medical records based on doctor-patient dialogue as taught by Wang, with the motivation of facilitating patient care (Wang Para. 0002).
Tang/ Wang does not explicitly teach, however Varshney teaches
randomly selecting a disease from a medical knowledge graph and capturing a plurality of symptoms that are mapped to the selected disease; ([Pg. 1 Abstract] The medical-specific knowledge graph contains broadly 3 types of medical-related information, including disease, symptom and laboratory test.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang, generation of medical records based on doctor-patient dialogue as taught by Wang, and incorporate a medical knowledge graph as taught by Varshney, with the motivation of improving access to healthcare services, improving patient treatment quality, and lowering medical expenses (Varshney Abstract).
REGARDING CLAIM 9
Claim(s) 9 is/are analogous to Claim(s) 2, thus Claim(s) 9 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 2.
REGARDING CLAIM 10
Claim(s) 10 is/are analogous to Claim(s) 4, thus Claim(s) 10 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 4.
REGARDING CLAIM 11
Tang/ Wang/ Varshney teach the system of claim 10, Tang further teaches wherein the format of the prepared input data is: [TOPICS] <symptom values> [ENTITIES] <entity values> [SPEAKER] <doctor/patient> [TURN LENGTH] <length of the turn>[REMAINING_TURNS] <number of turns left> [CONTEXT] <previous dialogues if available>. ([Pg. 2 2.1 Task Definition] The given inputs are in the form of a text sequence X = {x1, x2, ..., xn}, which consists of a patient’s question alongside the collection of prior dialogue turns between the doctor and the patient. ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.))
REGARDING CLAIM 14
Tang/ Wang/ Varshney teach the system of claim 8, Tang further teaches wherein the machine learning model has a transformer- encoder-decoder architecture. ([Pg. 2 2.3 Response Generation] We employ BART (encoder-decoder structure) as the base model.)
REGARDING CLAIM 15
Tang teach a computer-program product tangibly embodied in one or more non-transitory machine-readable media, including instructions configured to cause one or more data processors to perform the following operations:
obtaining an original dataset for training a machine learning model to generate a synthetic doctor-patient conversation, the original dataset including data in the form of textual dialogue associated with each of a plurality of individual doctor-patient conversations; ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.)
generating prepared input data by arranging the input data in an annotated turn-by-turn conversation format using an input data preparation algorithm having various control parameters; ([Pg. 2 2.2. Terminological Knowledge Enhancement] Obtain a corpus containing nearly 250,000 doctor-patient conversation pairs with 3,600M tokens. We set up our experiment based on this terminology-enhanced dataset, whose statistics
are shown in Table 1. Table 1 showcases data statistics of the Medical English Dialogue Corpus with annotated medical terminologies (abbr. Terms).)
training the machine learning model using the prepared input data; ([Pg. 2 2.3 Response Generation] We employ BART (encoder-decoder structure) as the base model. [Pg. 3 2.4 and 2.5] An auxiliary classification task is introduced to train the encoder to identify terminology-related tokens. Both aforementioned task objectives are combined to train the language model, and the overall loss function Loverall is minimized during model fine-tuning:)
and causing the trained machine learning model to generate a turn-by-turn synthetic doctor- patient conversation according to a conversation generation control algorithm having various control parameters, by inputting the plurality of symptoms to the machine learning model as a first control parameter of the conversation generation control algorithm and randomly assigning a first speaker. ([Pg. 2 2.1 Task Definition] The given inputs are in the form of a text sequence X = {x1, x2, ..., xn}, which consists of a patient’s question alongside the collection of prior dialogue turns between the doctor and the patient. The goal of the task is to generate a response Y = {y1, y2, ..., ym} (as a doctor) by modeling the conditional probability distribution P(Y |X). ([Pg.1 Introduction] Fig. 1. A medical dialogue example from the provided dataset. The phrases in red denote terminology-related expressions that have been automatically annotated by our framework, which include Symptoms (sticks out), Examines (physical exam), Diseased parts (spine), and others.))
Tang does not explicitly teach, however Wang teaches
constructing input data by performing a named entity recognition operation on the dialogue of the original dataset that captures and categorizes named medical entities present in the dialogue; ([Para. 0026] The output of section extractor 231, sections of conversation 240, is the input to named entity recognition engine 241, and thus, for illustrative purposes, section extractor 231 is shown as overlapping with, or intersecting, named entity recognition engine 241, precisely at sections of conversation 240. Named entity recognition engine 241 is configured to recognize various types of words or phrases of relevance to generating a medical record according to the hard medical record format, such as, for example, a symptom, a diagnosis, a drug, or an occupation, as shown in FIG. 4 at block 450, for example. The output of named entity recognition engine 241 is medical entities 250.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang and incorporate generation of medical records based on doctor-patient dialogue as taught by Wang, with the motivation of facilitating patient care (Wang Para. 0002).
Tang/ Wang does not explicitly teach, however Varshney teaches
randomly selecting a disease from a medical knowledge graph and capturing a plurality of symptoms that are mapped to the selected disease; ([Pg. 1 Abstract] The medical-specific knowledge graph contains broadly 3 types of medical-related information, including disease, symptom and laboratory test.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang, generation of medical records based on doctor-patient dialogue as taught by Wang, and incorporate a medical knowledge graph as taught by Varshney, with the motivation of improving access to healthcare services, improving patient treatment quality, and lowering medical expenses (Varshney Abstract).
REGARDING CLAIM 16
Claim(s) 16 is/are analogous to Claim(s) 2, thus Claim(s) 16 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 2.
REGARDING CLAIM 17
Claim(s) 17 is/are analogous to Claim(s) 4, thus Claim(s) 17 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 4.
REGARDING CLAIM 18
Claim(s) 18 is/are analogous to Claim(s) 11, thus Claim(s) 18 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 11.
REGARDING CLAIM 20
Claim(s) 20 is/are analogous to Claim(s) 14, thus Claim(s) 20 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 14.
Claim(s) 6, 12, and 19 is/are rejected under 35 U.S.C. 103 as being unpatentable over Tang (Terminology-aware Medical Dialogue Generation) in view of Wang (US 20210082585 A1) in view of Varshney (Knowledge graph assisted end-to-end medical dialog generation) in view of Erdenee (US 20230077528 A1).
REGARDING CLAIM 6
Tang/ Wang/ Varshney teach the computer-implemented method of claim 4, Erdenee wherein training the machine learning model includes multistage finetuning of the machine learning model, the multistage finetuning comprising:
first-fold finetuning using a medical question answering dataset; ([Abstract] A training method of a conversation model according to various example embodiments of the present disclosure may include identifying a first context, identifying a first response set corresponding to the first context based on a first model, identifying a response subset selected from the first response set based on a gold response corresponding to the first context)
following first-fold finetuning, second-fold finetuning guided by the input data preparation algorithm and the control parameters thereof and using a dataset including data in the form of textual dialogue associated with each of a plurality of individual non-medical specific conversations; ([Para. 0053] The electronic apparatus according to the example embodiments includes a conversation response training model that connects the retrieval model and the generation model, and the electronic apparatus according to the example embodiments may perform some or all of the operations of training the exemplar-based generation model. In training the exemplar-based generation model, the electronic apparatus according to the example embodiments may use a query for selecting an exemplar by using a gold response as well as context. )
and following second-fold finetuning, final-fold finetuning guided by the input data preparation algorithm and the control parameters thereof and using a dataset including data in the form of textual dialogue associated with each of a plurality of individual medical-specific conversations, wherein the symptoms/topics control parameter of the input data preparation algorithm is composed of keywords previously derived from a dataset including data in the form of textual dialogue associated with each of a plurality of individual non-medical conversations. ([Para. 0122] In the process of training using the training set data, the generation model according to the example embodiments may be trained in various ways according to the characteristics of the exemplar(s) provided from the retrieval model, the contents of the exemplar(s) and other data provided. Referring to a first exemplar 302a, a second exemplar 302b, and a third exemplar 302c in FIG. 3, the generation model according to the example embodiments may generate a response irrelevant to the context shown in operation 300 and the selected exemplar(s) shown in operation 301 as in the first exemplar 302a, or may generate a response that is excessively identical to the selected exemplar shown in operation 301 as in the second exemplar 302b.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang, generation of medical records based on doctor-patient dialogue as taught by Wang, and incorporate a medical knowledge graph as taught by Varshney, and incorporate A training method of a conversation model as taught by Erdenee, with the motivation of generating conversation information using an exemplar-based generation model (Erdenee Para. 0002).
REGARDING CLAIM 12
Claim(s) 12 is/are analogous to Claim(s) 6, thus Claim(s) 12 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 6.
REGARDING CLAIM 19
Claim(s) 19 is/are analogous to Claim(s) 6, thus Claim(s) 19 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 6.
Claim(s) 7 and 13 is/are rejected under 35 U.S.C. 103 as being unpatentable over Tang (Terminology-aware Medical Dialogue Generation) in view of Wang (US 20210082585 A1) in view of Varshney (Knowledge graph assisted end-to-end medical dialog generation) in view of Gunasekara (US 20230079879 A1).
REGARDING CLAIM 7
Tang/ Wang/ Varshney teach the computer-implemented method of claim 1, however Gunasekara teaches further comprising evaluating performance of the trained machine learning model by analyzing the generated turn-by-turn synthetic doctor-patient conversation using a metric selected from the group consisting of recall- oriented understudy for gisting evaluation, n-gram diversity score, unique n-gram count, and combinations thereof. ([Para. 0030] Once the summary generator 202 generates an output summary for the generated conversation 208, the reward model 204 can compare the output summary with the ground truth summary 206 that was used to ground the conversation generation of the conversation generator 102. In various examples, the reward model 204 can compare the ground truth summary 206 and generated summaries from the summary generator 202, and uses an output score as a reward. For example, the reward model 204 may use a Recall-Oriented Understudy for Gisting Evaluation (ROUGE)-2 F.sub.1 score as the reward. ROUGE-2 evaluates the overlap of bigrams between a system and reference summaries, and ROUGE F-1 is the harmonic mean of ROUGE-Recall.)
Therefore, it would be prima facie obvious to one of ordinary skill in the art, at the time of filing, to modify the method of medical dialogue generation as taught by Tang, generation of medical records based on doctor-patient dialogue as taught by Wang, and incorporate a medical knowledge graph as taught by Varshney, and incorporate training summary-grounded conversation generator, with the motivation of automatic generation of conversation (Gunasekara Para. Para. 0001)
REGARDING CLAIM 13
Claim(s) 13 is/are analogous to Claim(s) 7, thus Claim(s) 13 is/are similarly analyzed
and rejected in a manner consistent with the rejection of Claim(s) 7.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Strader et al (US 20190121532 A1), which discloses generating a transcript of a conversation between a patient and a healthcare practitioner
Cheng et al, Named Entity Recognition for Medical Dialogue Based on BERT and Adversarial Training, which proposes a BERT-based medical dialogue named entity recognition method BERT-BiLSTM-CRF-ADV. First, we use the BERT pre-training model to obtain word vectors with rich semantic information, then send the word vectors to BiLSTM to extract features, and finally input them into CRF for restriction correction and output. During the training process, we use adversarial training to improve the model performance. This paper applies the proposed method to the medical field to perform named entity recognition on medical dialogue materials.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Patricia K Edouard whose telephone number is (571)272-6084. The examiner can normally be reached Monday - Friday 7:30 AM - 5:00 PM.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Fonya M Long can be reached at 571-270-5096. 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.
/P.K.E./Examiner, Art Unit 3682
/FONYA M LONG/Supervisory Patent Examiner, Art Unit 3682
Prosecution Insights