A method and a generating device for generating an electronic medical record based on a doctor-patient conversation

Abstract

The application discloses a method and a generating device for generating an electronic medical record based on a doctor-patient conversation, relates to the technical field of medical information, and is used for improving the accuracy of speech recognition when generating an electronic medical record based on a doctor-patient conversation.In the method, the generating device automatically distinguishes subjective self-reports of patients and empirical symptoms supported by objective evidence based on the doctor-patient conversation, and fills them into different columns of the medical record respectively, and meanwhile, the empirical symptoms are sorted according to the strength of the objective evidence, so that the accuracy of speech recognition is improved when generating an electronic medical record based on a doctor-patient conversation.

Description

Technical Field

[0001] This application relates to the field of medical information technology, and in particular to a method and device for generating electronic medical records based on doctor-patient dialogue. Background Technology

[0002] With the deepening of medical informatization, electronic medical records have become an important carrier for clinical diagnosis and treatment in hospitals.

[0003] In busy outpatient clinics, to alleviate the burden of medical record entry for doctors and improve diagnostic and treatment efficiency, an AI-based assisted medical record entry technology has been developed. Specifically, this technology acquires the doctor-patient dialogue audio, uses automatic speech recognition technology to transcribe it into dialogue text data, and extracts entities such as symptoms and medical history. Simultaneously, it uses a document scanner or OCR technology to convert the patient's examination and test reports into report text data. After acquiring this text information, keyword matching, predefined rule templates, or natural language generation models are used to fill in the key information extracted from the dialogue and report text data into the corresponding columns of the medical record, such as the chief complaint, present illness, and auxiliary examinations, ultimately generating an electronic medical record for the doctor's confirmation.

[0004] However, when speech recognition is affected by accent interference or when the patient's description does not match the examination results, the relevant technology has difficulty correcting the speech text, which may result in the inclusion of incorrect oral statements or contradictory information in the medical record, leading to low accuracy of speech recognition. Summary of the Invention

[0005] This application provides a method and device for generating electronic medical records based on doctor-patient dialogue, which improves the accuracy of speech recognition when generating electronic medical records based on doctor-patient dialogue.

[0006] Firstly, a method for generating electronic medical records based on doctor-patient dialogue is provided, characterized by being applied to a generation device. The method includes: performing acoustic feature matching on the doctor-patient dialogue speech stream based on a preset medical terminology feature library to obtain multiple candidate symptom words containing acoustic confidence scores, where the acoustic confidence score is the similarity between the sound waveform of the candidate symptom word and the standard pronunciation features in the preset medical terminology feature library; searching for associated indicator items with a pathological causal relationship with the candidate symptom words in an objective indicator set based on a preset subjective-objective association mapping table to obtain multiple anchor pairs containing candidate symptom words and associated indicator items, where the objective indicator set is a set of objective indicators extracted from the patient's examination report, containing item names, test values, and reference range intervals; for each anchor pair, based on the detection of associated indicator items... The objective evidence support coefficient is calculated based on the degree of deviation of the measured value from the reference range of the associated indicator item. If the acoustic confidence of the candidate symptom word is less than the preset first threshold, it is determined whether the largest objective evidence support coefficient among all anchor pairs corresponding to the candidate symptom word is greater than the preset second threshold. If so, the candidate symptom word is replaced with the standard medical terminology corresponding to the associated indicator item in the preset subjective-objective association mapping table. The replaced candidate symptom word is marked as empirical symptom data. If not, the candidate symptom word is marked as symptom data to be verified. The empirical symptom data is filled into the objective corroboration item of the present medical history column, and the symptom data to be verified is filled into the self-report sub-item of the chief complaint column. The data in the objective corroboration item are prioritized based on the objective evidence support coefficient to generate an electronic medical record.

[0007] By adopting the above technical solution, the device automatically distinguishes between the patient's subjective self-report and the empirical symptoms supported by objective evidence based on the doctor-patient dialogue, and fills them into different columns of the medical record respectively. At the same time, the empirical symptoms are sorted according to the strength of their objective evidence, thereby improving the accuracy of speech recognition when generating electronic medical records based on doctor-patient dialogue.

[0008] In conjunction with some embodiments of the first aspect, in some embodiments, a preset subjective-objective association mapping table is included, specifically: the preset subjective-objective association mapping table is a key-value pair data table with the terminology of standard symptoms as the key value and the name of one or more clinical test items that cause standard symptoms as the index value.

[0009] By adopting the above technical solution, and by concretizing the subjective-objective correlation mapping table into a key-value pair data table with standard symptoms as keys and clinical test items as indexes, a clear structure is provided for storing and querying the pathological relationship between symptoms and objective indicators. This structured design facilitates the rapid construction, maintenance, and expansion of the medical knowledge base.

[0010] In conjunction with some embodiments of the first aspect, in some embodiments, the step of determining whether the largest objective evidence support coefficient among all anchor pairs corresponding to the candidate symptom word is greater than a preset second threshold specifically includes: a preset subjective-objective association mapping table also stores a preset pathological association weight value for each key-value pair, the pathological association weight value representing the probability that an abnormality in the clinical test item corresponding to the index value leads to the occurrence of the standard symptom corresponding to the key value; for each anchor pair, multiplying the objective evidence support coefficient corresponding to the anchor pair with the preset pathological association weight value corresponding to the preset subjective-objective association mapping table to obtain a comprehensive pathological explanatory index; and determining whether the largest comprehensive pathological explanatory index among all anchor pairs corresponding to the candidate symptom word is greater than a preset second threshold.

[0011] By adopting the above technical solution, the generation device further introduces a pathological association weight value that characterizes the strength of pathological association on the basis of the objective evidence support coefficient. The objective evidence support coefficient and the pathological association weight value are combined into a comprehensive pathological interpretation index, so that the generation device no longer relies solely on the degree of abnormality of the test results, but comprehensively considers the degree of abnormality and the strength of association, thereby making the judgment logic closer to the thinking mode of clinicians.

[0012] In conjunction with some embodiments of the first aspect, in some embodiments, the step of replacing candidate symptom words with standard medical terms corresponding to the associated index items in a preset subjective-objective association mapping table specifically includes: the preset subjective-objective association mapping table also stores a degree modifier classification table associated with clinical test items, the degree modifier classification table defining degree modifiers corresponding to comprehensive pathological explanatory indices of different numerical ranges; based on the comprehensive pathological explanatory index of the target anchor pair corresponding to the maximum comprehensive pathological explanatory index, searching for matching target degree modifiers in the degree modifier classification table; semantically concatenating the target degree modifiers and the standard medical terms corresponding to the target anchor pairs in the preset subjective-objective association mapping table to generate combined empirical semantic terms; and replacing candidate symptom words with combined empirical semantic terms.

[0013] By adopting the above technical solution, the generation device introduces a degree modifier grading table to transform the quantified comprehensive pathological interpretation index into qualitative descriptions such as mild, moderate, and severe, which are easier for clinicians to understand. These degree modifiers are then semantically concatenated with standard medical terminology, enriching the information of candidate symptom terms and thus providing doctors with more refined electronic medical records.

[0014] In conjunction with some embodiments of the first aspect, in some embodiments, the step of calculating the objective evidence support coefficient based on the deviation of the detected value of the associated indicator item from the reference range interval of the associated indicator item specifically includes: when the detected value of the associated indicator item is numerical, the reference range interval includes an upper threshold and a lower threshold; if the detected value of the associated indicator item is greater than the upper threshold, then the difference between the detected value and the upper threshold is calculated; the ratio of the difference to the upper threshold is determined as the objective evidence support coefficient; if the detected value of the associated indicator item is less than the lower threshold, then the difference between the lower threshold and the detected value is calculated; the ratio of the difference to the lower threshold is determined as the objective evidence support coefficient; if the detected value is between the upper threshold and the lower threshold, then the objective evidence support coefficient is determined to be zero.

[0015] By adopting the above technical solution, the generating device calculates the deviation ratio of the detected value from the reference range threshold, thereby normalizing and quantifying the degree of abnormality of different detection items, thus eliminating the influence of different item dimensions and units.

[0016] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of including an upper threshold and a lower threshold in the reference range when the detected value of the associated indicator item is numerical, the method further includes: when the detected value of the associated indicator item is qualitative textual, obtaining a preset qualitative quantization mapping table, the preset quantization mapping table storing the mapping relationship between different qualitative textual results and preset quantization coefficient values; searching for a target preset quantization coefficient value that matches the detected value in the qualitative quantization mapping table; and determining the target preset quantization coefficient value as the objective evidence support coefficient.

[0017] By adopting the above technical solution, the generation device supplements the processing method for qualitative text detection results. Through a preset qualitative quantification mapping table, these non-numerical text results are transformed into standardized quantification coefficient values, thereby incorporating them into the unified calculation framework of objective evidence support coefficients.

[0018] In conjunction with some embodiments of the first aspect, in some embodiments, after the step of marking the replaced candidate symptom words as empirical symptom data, the method further includes: obtaining the target item name and target detection value of the associated indicator item in the anchor pair on which the empirical symptom data is based; encapsulating the target item name and target detection value into an evidence tracing label according to a preset tracing format; and appending the evidence tracing label to the end of the empirical symptom data.

[0019] By adopting the above technical solution, the generation device further improves the rigor of electronic medical records and increases the efficiency of doctors' review of electronic medical records by attaching an evidence traceability label containing specific indicators and values ​​to each empirical symptom data confirmed by objective evidence.

[0020] In a second aspect, embodiments of this application provide a generation device, which includes: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is used to store computer program code, the computer program code including computer instructions, wherein the one or more processors invoke the computer instructions to cause the generation device to perform the method as described in the first aspect and any possible implementation thereof.

[0021] Thirdly, embodiments of this application provide a computer program product containing instructions that, when the computer program product is run on a generating device, cause the generating device to perform the method described in the first aspect and any possible implementation thereof.

[0022] Fourthly, embodiments of this application provide a computer-readable storage medium including instructions that, when executed on a generating device, cause the generating device to perform the method described in the first aspect and any possible implementation thereof.

[0023] Understandably, the generating apparatus provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the methods provided in the embodiments of this application. Therefore, the beneficial effects they can achieve can be referred to the beneficial effects in the corresponding methods, and will not be repeated here.

[0024] One or more technical solutions provided in the embodiments of this application have at least the following technical effects or advantages:

[0025] 1. The device automatically distinguishes between the patient's subjective self-report and the empirical symptoms supported by objective evidence based on the doctor-patient dialogue, and fills them into different columns of the medical record. At the same time, the empirical symptoms are sorted according to the strength of their objective evidence, thereby improving the accuracy of speech recognition when generating electronic medical records based on doctor-patient dialogue.

[0026] 2. Based on the objective evidence support coefficient, the generation device further introduces a pathological association weight value to characterize the strength of pathological association. It combines the objective evidence support coefficient and the pathological association weight value into a comprehensive pathological interpretation index. This makes the generation device's judgment of symptoms no longer solely dependent on the degree of abnormality of the test results, but comprehensively considers both the degree of abnormality and the strength of association, thus making the judgment logic closer to the thinking mode of clinicians.

[0027] 3. The generation device, by introducing a severity modifier grading table, transforms the quantified comprehensive pathological interpretation index into qualitative descriptions such as mild, moderate, and severe, which are easier for clinicians to understand. These severity modifiers are then semantically concatenated with standard medical terminology, enriching the information of candidate symptom terms and thus providing doctors with more refined electronic medical records. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating a method for generating electronic medical records based on doctor-patient dialogue in an embodiment of this application.

[0029] Figure 2 This is another flowchart illustrating a method for generating electronic medical records based on doctor-patient dialogue in an embodiment of this application.

[0030] Figure 3 This is a schematic diagram of the physical device structure of the generating device in the embodiments of this application. Detailed Implementation

[0031] The terminology used in the following embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. As used in the specification and appended claims of this application, the singular expressions “a,” “an,” “the,” “the,” “the,” and “this” are intended to include the plural expressions as well, unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this application refers to and includes any or all possible combinations of one or more of the listed items.

[0032] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as implying or suggesting relative importance or implicitly indicating the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature, and in the description of the embodiments of this application, unless otherwise stated, "multiple" means two or more.

[0033] This application provides a method and device for generating electronic medical records based on doctor-patient dialogue, which improves the accuracy of speech recognition when generating electronic medical records based on doctor-patient dialogue.

[0034] Please see Figure 1 This is a flowchart illustrating a method for generating electronic medical records based on doctor-patient dialogue in an embodiment of this application.

[0035] S101. Based on the preset medical terminology feature library, perform acoustic feature matching on the doctor-patient dialogue voice stream to obtain multiple candidate symptom words containing acoustic confidence.

[0036] The doctor-patient dialogue voice stream refers to the continuous audio data of the conversation between the doctor and the patient, collected or recorded in real time during the diagnosis and treatment process. The pre-built medical terminology feature library is a pre-constructed database that stores a large number of acoustic feature models of the standard pronunciations of standard medical terms (such as dizziness, fatigue, and low hemoglobin), including Mel-frequency cepstral coefficients and phoneme models. Acoustic feature matching refers to the process of comparing and calculating the acoustic features of the input voice stream segment with the standard pronunciation features stored in the feature library. Acoustic confidence is a quantitative value, usually between 0 and 1, used to represent the reliability of judging a voice segment as a specific medical term; a higher value indicates higher reliability. Candidate symptom words refer to one or more medical terminology texts that may correspond to the voice segment, output by the acoustic feature matching process; each candidate word is accompanied by its corresponding acoustic confidence.

[0037] Specifically, when the generation device starts working, it continuously receives and processes the doctor-patient dialogue voice stream from audio acquisition devices such as microphones. The generation device segments this continuous voice stream into short audio frames and extracts acoustic features from these frames. Subsequently, it calls a preset medical terminology feature library and matches the extracted acoustic features with the feature models of various standard medical terms stored in the library using algorithms such as dynamic time warping or Hidden Markov Models. The matching process calculates a similarity score, which, after normalization, becomes the acoustic confidence score. For a symptom that the patient pronounces unclearly, such as a sound that sounds between "tóu yūn" and "tóu téng", the generation device, after comparing it with the preset medical terminology feature library, may output a set of candidate symptom words, such as {(dizziness, 0.75), (headache, 0.68)}. Both words are standard terms in the feature library, but due to their acoustic similarity, they are both identified as possible candidates and given different acoustic confidence levels. These candidate words, along with their confidence levels, are sent to subsequent processing steps.

[0038] In some embodiments, acoustic feature matching can be performed in several ways: Optionally, the generation device can employ a traditional speech recognition framework based on a Gaussian Mixture Model-Hidden Markov Model. First, MFCC features of the speech stream are extracted. Second, the acoustic feature distribution of each phoneme is modeled using a Gaussian Mixture Model (GMM). Finally, the MFCC sequence is decoded at the phoneme level using an Hidden Markov Model (HMM) (whose dictionary is defined by a pre-defined medical terminology feature library), and the word sequence with the highest probability is found as a candidate symptom word using the Viterbi algorithm. The cumulative probability of this path is then used as the acoustic confidence score. Optionally, the generation device can also employ an end-to-end speech recognition model based on deep learning, such as an encoder-decoder model based on connection-time classification or attention mechanisms. This model directly maps the input acoustic feature sequence to the candidate symptom word sequence, and the probability values ​​of the model's output layer (such as a Softmax layer) can be directly used as the acoustic confidence score. This approach typically has higher recognition accuracy, especially when dealing with accents and noise.

[0039] S102. Based on the preset subjective-objective correlation mapping table, search for the associated indicator items in the objective indicator set that have a pathological causal relationship with the candidate symptom words, and obtain multiple anchor pairs containing candidate symptom words and associated indicator items.

[0040] The pre-defined subjective-objective correlation mapping table is a structured data table that uses the terminology of standard symptoms as keys and the names of one or more clinical test items that may cause the symptom as indexes, establishing a pathological association between subjective symptoms and objective test indicators. For example, the key value "anemia" may correspond to the indexes of hemoglobin and red blood cell count. The objective indicator set refers to structured information extracted from various patient examination and test reports (such as complete blood count, urinalysis, and imaging reports) through optical character recognition or data interfaces. This set typically contains multiple indicator items, each with an item name, test value, and reference range. Associated indicator items are specific test items found in the objective indicator set and associated with a candidate symptom term in the subjective-objective correlation mapping table. An anchored pair represents a data pair consisting of a candidate symptom term and its associated objective indicator item, anchoring a potentially uncertain subjective expression to a definite objective test result.

[0041] Specifically, after obtaining candidate symptom terms with acoustic confidence levels, the generation device iterates through each candidate symptom term. For a candidate symptom term, such as fatigue, the generation device first searches in a pre-defined subjective-objective association mapping table. The mapping table may return the names of multiple clinical test items related to fatigue, such as complete blood count - hemoglobin and electrolytes - serum potassium. Next, the generation device searches in the pre-parsed set of objective indicators based on these returned item names. If an item named hemoglobin is found in the set of objective indicators, and its test value and reference range both exist, then the generation device creates an anchor pair with the content {candidate symptom term: fatigue, associated indicator item: {item name: hemoglobin, test value: 95g / L, reference range: 110-150g / L}}. If a candidate symptom term is associated with multiple objective indicators, and these indicators are all found in the report, multiple anchor pairs will be generated.

[0042] S103. For each anchor pair, calculate the objective evidence support coefficient based on the degree of deviation of the detected value of the associated indicator relative to the reference range of the associated indicator.

[0043] Specifically, for each anchor pair obtained in S102, the generating device extracts the associated indicator information for the anchor pair, including the test value and reference range. Then, it determines whether the test value is within the reference range. If the test value falls within the range, it indicates that the indicator is normal, and its objective evidence support for the related symptoms is negligible; in this case, the objective evidence support coefficient is calculated as zero. If the test value exceeds the upper or lower limit of the reference range, the generating device calculates a non-zero coefficient value based on the degree of deviation. For example, for an indicator with a reference range of [3.5, 5.5], a test value of 6.5 deviates more than a test value of 5.6; therefore, the former has a higher objective evidence support coefficient, which reflects the severity of the abnormality in the laboratory test result.

[0044] In some embodiments, the calculation of the objective evidence support coefficient in this step can be implemented in several ways: Optionally, when the detection value of the associated indicator is numerical, the generating device first parses the upper and lower thresholds of the reference range. If the detection value is greater than the upper threshold, the difference (detection value - upper threshold) is calculated, and the ratio of this difference to the upper threshold is used as the objective evidence support coefficient. If the detection value is less than the lower threshold, the difference (lower threshold - detection value) is calculated, and the ratio of this difference to the lower threshold is used as the objective evidence support coefficient. This normalized calculation method can eliminate the influence of different indicator dimensions. Optionally, when the detection value of the associated indicator is qualitative text (such as negative, positive, and ++), the generating device queries a preset qualitative-quantitative mapping table. This mapping table stores the correspondence between different qualitative text results and preset quantitative coefficient values; for example, negative corresponds to 0, weak positive or + corresponds to 0.5, and positive or ++ corresponds to 1.0. The generating device finds the text that matches the current detection value in the table and directly uses its corresponding preset quantitative coefficient value as the objective evidence support coefficient.

[0045] S104. If the acoustic confidence of a candidate symptom word is less than a preset first threshold, determine whether the largest objective evidence support coefficient among all anchor pairs corresponding to the candidate symptom word is greater than a preset second threshold.

[0046] The preset first threshold is a judgment threshold set for acoustic confidence, such as 0.7, used to distinguish between a clear and reliable speech recognition result and a vague and uncertain one. The preset second threshold is a judgment threshold set for the objective evidence support coefficient, such as 0.2, used to determine whether the degree of abnormality of an objective examination result is sufficient as strong evidence to correct an uncertain subjective statement.

[0047] Specifically, the generation device first checks the acoustic confidence level of a candidate symptom word. If the confidence level is greater than or equal to a preset first threshold (e.g., ≥0.7), it indicates that the speech recognition result is relatively reliable, and the generation device skips this judgment step, directly treating the candidate symptom word as reliable information to be processed. Conversely, if the acoustic confidence level is less than the first threshold (e.g., <0.7), it indicates that this may be a blurry recognition caused by accent, speech rate, or noise, requiring additional evidence for confirmation. At this time, the generation device examines all anchor pairs associated with the candidate symptom word and finds the largest objective evidence support coefficient. Then, it compares this largest coefficient value with a preset second threshold. This process is analogous to a decision: when something is unclear, check the lab report; if the lab report shows obvious abnormalities and these abnormalities are pathologically related to the unclear symptom heard, then accept that symptom.

[0048] In some embodiments, the threshold judgment and decision logic in this step can be implemented in several ways: Optionally, a fixed dual-threshold judgment can be used. The first threshold (e.g., 0.75) and the second threshold (e.g., 0.15) are hard-coded in the generation device. The entire judgment process follows Boolean logic with an acoustic confidence level < 0.75 and a maximum objective evidence support coefficient > 0.15. This approach is simple and efficient, and suitable for scenarios with clearly defined business rules. Optionally, a dynamic or adaptive threshold judgment can also be used. The first and second thresholds are not fixed, but can be dynamically adjusted according to the dialogue context, doctor level, or department characteristics. For example, in a follow-up visit scenario, if the patient is known to have a certain chronic disease, the second threshold can be appropriately lowered, as slight fluctuations in the indicator may also be meaningful. Alternatively, the threshold can be learned and optimized by a machine learning model based on historically confirmed medical record data to achieve a higher overall accuracy.

[0049] S105. Replace candidate symptom words with standard medical terms corresponding to the associated indicator items in the preset subjective-objective association mapping table.

[0050] Specifically, the generating device executes this step when the judgment result of step S104 is yes (i.e., low acoustic confidence but strong objective evidence). This means that although the patient's unclear pronunciation leads to multiple possible candidate symptom words acoustically, one of the candidate words implies that it is corroborated by the objective examination results. The generating device will locate the anchor pair that provides the largest objective evidence support coefficient. Then, based on the associated index item in this anchor pair, it reversely searches for the corresponding standard symptom term (i.e., the key in the table) in a preset subjective-objective association mapping table, and the generating device replaces the candidate symptom word with the standard symptom term. For example, suppose the output of S101 is two acoustically similar candidate symptom words: {(anemia, 0.65), (fatigue, 0.62)}, both of which have low acoustic confidence. In subsequent steps, the generating device finds that the patient's test report shows a low hemoglobin level, which provides a very high objective evidence support coefficient for anemia, while indicators directly related to fatigue (such as electrolytes) may be normal. Therefore, the generating device selects anemia as the candidate symptom word to replace it, and determines that the candidate symptom word is anemia.

[0051] S106. Mark the replaced candidate symptom words as empirical symptom data.

[0052] Specifically, in step S105, once a vague candidate symptom term is replaced by a standard medical term supported by objective evidence, the generating device marks the new candidate symptom term as empirical symptom data.

[0053] In some embodiments, to improve the rigor and traceability of electronic medical records, the generating device performs an operation to attach an evidence tracing tag to support the decision made in this step. The generating device acquires the anchor pair on which the empirical symptom data was based and extracts the target item name (e.g., hemoglobin) and target detection value (e.g., 95 g / L) of the associated indicator from the anchor pair. Then, according to a preset tracing format, such as [Tracing: Hemoglobin = 95 g / L], this information is encapsulated into a structured evidence tracing tag. Finally, this evidence tracing tag is appended to the end of the empirical symptom data. For example, the final generated data item might be anemia [Tracing: Hemoglobin = 95 g / L]. This not only improves the certainty of the information but also provides doctors with a direct and quickly verifiable link to objective evidence during review.

[0054] S107. Mark candidate symptom words as symptom data to be verified.

[0055] Among them, unverified symptom data is another data label or type. In contrast to empirical symptom data, it is used to identify symptom information that lacks objective evidence or is uncertain in speech recognition itself, indicating that its credibility is low and requires manual confirmation by a doctor.

[0056] Specifically, this step is another branch of the S104 judgment. When a candidate symptom word with low acoustic confidence fails to exceed the preset second threshold in the maximum objective evidence support coefficient among all its corresponding anchor pairs after the judgment in step S104 (i.e., the judgment result is negative), the generating device executes this step. This indicates that the sound is unclear and no corresponding abnormality is found on the test report, making it difficult for the generating device to confirm the authenticity of the symptom. Therefore, the generating device will mark this original, unmodified candidate symptom word (e.g., dizziness with an acoustic confidence of 0.6, but all relevant test indicators are normal) as symptom data to be verified. In addition, for candidate symptom words with a high acoustic confidence (greater than or equal to the first threshold), in the logic of this embodiment, they can also be directly marked as symptom data to be verified because although they are clear, they have not been verified by objective indicators.

[0057] S108. Fill in the empirical symptom data into the objective corroboration item of the present medical history column, and fill in the symptom data to be verified into the self-report sub-item of the chief complaint column.

[0058] The "Present Illness" section in the electronic medical record documents the onset, progression, and treatment of the patient's current illness. The "Objective Supporting Evidence" section is a sub-item under the "Present Illness" section, specifically used to record descriptions of the condition supported by objective examination results. The "Chief Complaint" section records the primary reason and symptoms for the patient's current visit. The "Personal Report" sub-item is a sub-item under the "Chief Complaint" section, used to record the patient's own statements, the content of which has not been objectively verified.

[0059] Specifically, after classifying and labeling all candidate symptom terms, the device enters the medical record text generation stage. It iterates through all information marked as empirical symptom data and fills this information into the "Objective Evidence" section of the present illness chapter in the electronic medical record template according to a preset format. Simultaneously, it iterates through all information marked as symptom data to be verified and fills this information (e.g., dizziness identified with high confidence or abdominal pain with low confidence but no evidence) into the "Self-Report" sub-item of the chief complaint section. Through this differentiation, the generated medical record retains the patient's subjective feelings while highlighting key conditions supported by objective evidence, providing doctors with a clearly structured and focused information reference.

[0060] S109. Prioritize the data in the objective corroborating items based on the objective evidence support coefficient, and generate electronic medical records.

[0061] Specifically, after filling in all the empirical symptom data into the objective corroboration items, the generation device will perform a sorting operation to further improve the readability and clinical value of the medical record. Based on the objective evidence support coefficient obtained in calculation step S103 for each empirical symptom data, these data items are sorted in descending order. A higher objective evidence support coefficient indicates a more significant abnormality in the corresponding objective examination indicators, and the more pathologically important the symptom may be. For example, a symptom of polydipsia and polyuria supported by severely elevated blood sugar (its objective evidence support coefficient may be 1.5) will be ranked before a symptom of fever supported by a slightly elevated white blood cell count (its coefficient may be 0.3). After sorting, the content in the objective corroboration items will be presented in a logical order of severe abnormality - moderate abnormality - mild abnormality. Finally, the generation device integrates all the filled-in fields and outputs a complete, structured, and focused draft of the electronic medical record for physician review.

[0062] In the above embodiments, the generating device automatically distinguishes between the patient's subjective self-report and the empirical symptoms supported by objective evidence based on the doctor-patient dialogue, and fills them into different columns of the medical record respectively. At the same time, the empirical symptoms are sorted according to the strength of their objective evidence, thereby improving the accuracy of speech recognition when generating electronic medical records based on doctor-patient dialogue.

[0063] However, in complex clinical practice, the correlation between an abnormal indicator and a particular symptom varies in strength. For example, severely low hemoglobin levels are far more explanatory for fatigue than slightly elevated white blood cell counts. If an indicator with a weak pathological correlation to the symptom happens to show an extreme abnormality, while an indicator with a strong correlation shows only a slight abnormality, judging solely by the degree of abnormality may lead to decisions that do not fully align with clinical logic. To address this issue, embodiments of this application introduce a more refined assessment method.

[0064] Please see Figure 2 This is another flowchart illustrating a method for generating electronic medical records based on doctor-patient dialogue in an embodiment of this application.

[0065] S201. Based on the preset medical terminology feature library, perform acoustic feature matching on the doctor-patient dialogue voice stream to obtain multiple candidate symptom words containing acoustic confidence.

[0066] S202. Based on the preset subjective-objective correlation mapping table, search for the associated indicator items in the objective indicator set that have a pathological causal relationship with the candidate symptom words, and obtain multiple anchor pairs containing candidate symptom words and associated indicator items.

[0067] S203. For each anchor pair, calculate the objective evidence support coefficient based on the degree of deviation of the detected value of the associated indicator relative to the reference range of the associated indicator.

[0068] Step S201 is similar to step S101, step S202 is similar to step S102, and step S203 is similar to step S103, so they will not be described again here.

[0069] S204. The preset subjective-objective association mapping table also stores a preset pathological association weight value for each key-value pair.

[0070] In this context, a key-value pair refers to an entry in a pre-defined subjective-objective association mapping table, which is a combination of a standard symptom term (key) and its corresponding one or more clinical test item names (index values). The pre-defined pathological association weight value is a numerical value that is pre-set and stored in the mapping table. It characterizes the prior probability or medical strength of the association between an abnormality in a specific clinical test item and the occurrence of its associated standard symptom. For example, the weight value for hemoglobin abnormalities leading to anemia might be high (e.g., 0.9), while the weight value for platelet abnormalities leading to fatigue might be relatively low (e.g., 0.4).

[0071] Specifically, when constructing the preset subjective-objective correlation mapping table, in addition to establishing the connection relationship between standard symptom terms and clinical test item names, medical experts will also assign a weight to each pair of standard symptom term-clinical test item name associations. This weight is derived based on medical evidence, clinical guidelines, and a large amount of statistical data, reflecting how strongly abnormal a specific indicator can explain a specific symptom. When the generating device searches for associated indicator items in step S202, it will not only obtain the item name, test value, and reference range, but also the preset pathological correlation weight value.

[0072] S205. For each anchor pair, multiply the objective evidence support coefficient corresponding to the anchor pair with the preset pathological association weight value corresponding to the preset subjective-objective association mapping table to obtain the comprehensive pathological interpretation index.

[0073] Among them, the Comprehensive Pathological Interpretation Index is a new and more comprehensive assessment score that integrates information from two dimensions: how abnormal the examination results are (objective evidence support coefficient) and how strong the association between this abnormality and the symptoms is (pathological association weight value).

[0074] Specifically, following S204, for each anchor pair, the generating device obtains its objective evidence support coefficient and pathological association weight value. These two values ​​are then multiplied to obtain the overall pathological explanatory index for that anchor pair. For example, a candidate symptom term weakly associated with two indicators: indicator A has a support coefficient of 0.8 but a pathological association weight of only 0.3, resulting in an overall index of 0.24; indicator B has a support coefficient of 0.5 but a pathological association weight as high as 0.9, resulting in an overall index of 0.45. Through this calculation, the generating device can more accurately determine which objective evidence has the strongest explanatory power for the current symptom.

[0075] S206. If the acoustic confidence of a candidate symptom word is less than a preset first threshold, determine whether the maximum comprehensive pathological explanatory index among all anchor pairs corresponding to the candidate symptom word is greater than a preset second threshold.

[0076] The maximum comprehensive pathological explanatory index refers to the maximum value of the comprehensive pathological explanatory index calculated among all anchor pairs associated with a candidate symptom term.

[0077] Specifically, when the acoustic confidence level of a candidate symptom term is less than a preset first threshold, the system determines whether the maximum composite pathological explanatory index among all anchor pairs corresponding to the candidate symptom term is greater than a preset second threshold. Compared to S104, the judgment logic here is more accurate, no longer only considering the most abnormal test result, but also the test result with the strongest explanatory power. An indicator with moderate abnormality but high correlation with symptoms may have a composite index that exceeds that of an indicator with high abnormality but weak correlation with symptoms, thus making the judgment more consistent with clinical logic.

[0078] S207. Replace candidate symptom words with standard medical terms corresponding to the associated indicator items in the preset subjective-objective association mapping table.

[0079] In this embodiment, unlike S105, when the judgment result of step S206 is yes, the generating device locates the target anchor pair that contributed the largest comprehensive pathological explanatory index. Then, based on this largest comprehensive pathological explanatory index (e.g., 0.75), it queries a preset severity modifier grading table, which defines severity modifiers corresponding to comprehensive pathological explanatory indices in different numerical ranges. If the grading table defines the range [0.6, 0.9) as corresponding to moderate, then the found target severity modifier is moderate. Simultaneously, it retrieves the standard medical term (e.g., leukocytosis) corresponding to the target anchor pair from a preset subjective-objective association mapping table. Finally, the device semantically concatenates these two parts to generate a combined empirical semantic term, namely, moderate leukocytosis. This newly generated, more accurate term containing severity information will be used to replace the original, vague candidate symptom terms. This step makes the final generated medical record content more accurate and refined.

[0080] In some embodiments, the semantic concatenation process in this step can be implemented in several ways: Optionally, pre-concatenation can be used. The found target degree modifier is directly placed before the standard medical term, forming a modifier + term structure, for example, mild + anemia generates mild anemia. Optionally, template-based generation can also be used. The generation device presets some phrase templates that are more in line with natural language habits, such as [standard medical terminology], represented as [degree modifier] or [standard medical terminology] with visible [degree modifier]. The device fills the found words into the template to generate richer sentences, for example, leukocytosis is represented as moderate. This approach can improve the fluency and readability of the generated text.

[0081] S208. Mark the replaced candidate symptom words as empirical symptom data.

[0082] S209. Mark candidate symptom words as symptom data to be verified.

[0083] S210. Fill in the empirical symptom data into the objective corroboration item of the present medical history column, and fill in the symptom data to be verified into the self-report sub-item of the chief complaint column.

[0084] S211. Prioritize the data in the objective evidence items based on the objective evidence support coefficient, and generate electronic medical records.

[0085] Step S208 is similar to step S106, step S209 is similar to step S107, step S210 is similar to step S108, and step S211 is similar to step S109, so they will not be described again here.

[0086] In the above embodiments, the generation device introduces a pathological association weight value and multiplies it by an objective evidence support coefficient to obtain a comprehensive pathological explanation index. This allows the symptom assessment to simultaneously consider both the degree of abnormality of the indicator and the strength of the pathological association between the indicator and the symptom, making the decision-making process more scientific and closer to clinical thinking. Furthermore, by introducing a degree modifier grading table, the generation device transforms the quantified comprehensive pathological explanation index into qualitative descriptions such as mild, moderate, and severe, which are easier for clinicians to understand. Semantically concatenating these degree modifiers with standard medical terminology enriches the information of the candidate symptom words, thus providing doctors with more refined electronic medical records.

[0087] The above describes a method for generating electronic medical records based on doctor-patient dialogue in the embodiments of this application. The exemplary generation device 300 provided in the embodiments of this application is described below.

[0088] Figure 3 This is a schematic diagram of an exemplary hardware structure of the generation device 300 provided in an embodiment of this application. In some embodiments, the generation device 300 is a computer device. The computer device includes a processor, a memory, and a network interface connected via a system bus. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the computer device stores data. The network interface of the computer device is used to communicate with other external terminals or generation devices via a network connection. In some embodiments, the network interface can be a wired network interface; in some embodiments, the network interface can also be a wireless network interface. When the computer program is executed by the processor, it implements a method for generating electronic medical records based on doctor-patient dialogue in an embodiment of this application.

[0089] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0090] In some embodiments of this application, a computer-readable storage medium is also provided, including instructions that, when executed on the generating device 300, cause the generating device 300 to perform a method for generating electronic medical records based on doctor-patient dialogue according to an embodiment of this application.

[0091] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit it. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

[0092] As used in the above embodiments, depending on the context, the term "when..." can be interpreted as meaning "if...", "after...", "in response to determining...", or "in response to detecting...". Similarly, depending on the context, the phrase "when determining..." or "if (the stated condition or event) is interpreted as meaning "if determining...", "in response to determining...", "when (the stated condition or event) is detected", or "in response to detecting (the stated condition or event)".

[0093] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, production device, or data center to another website, computer, production device, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a production device or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state drive), etc.

[0094] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes described in the above method embodiments. The aforementioned storage medium includes various media capable of storing program code, such as ROM or random access memory (RAM), magnetic disks, or optical disks.

Claims

1. A method for generating electronic medical records based on doctor-patient dialogue, characterized in that, Applied to a generating device, the method includes: Based on a pre-set medical terminology feature library, acoustic feature matching is performed on the doctor-patient dialogue voice stream to obtain multiple candidate symptom words containing acoustic confidence. The acoustic confidence is the matching similarity between the sound waveform of the candidate symptom word and the standard pronunciation features in the pre-set medical terminology feature library. Based on a preset subjective-objective correlation mapping table, the system searches for related indicator items in the objective indicator set that have a pathological causal relationship with the candidate symptom words, thereby obtaining multiple anchor pairs containing the candidate symptom words and the related indicator items. The objective indicator set is a set of objective indicators extracted from the patient's examination report that includes the item name, test value, and reference range. For each anchor pair, an objective evidence support coefficient is calculated based on the degree of deviation of the detected value of the associated indicator relative to the reference range interval of the associated indicator. If the acoustic confidence of the candidate symptom word is less than a preset first threshold, determine whether the largest objective evidence support coefficient among all the anchor pairs corresponding to the candidate symptom word is greater than a preset second threshold. If so, the candidate symptom word is replaced with the standard medical term corresponding to the associated indicator item in the preset subjective-objective association mapping table; The replaced candidate symptom words are labeled as empirical symptom data; If not, then mark the candidate symptom words as symptom data to be verified; Enter the empirical symptom data into the objective corroboration field of the present medical history section, and enter the symptom data to be verified into the self-report sub-field of the chief complaint section; Based on the objective evidence support coefficient, the data in the objective corroboration items are prioritized and sorted to generate electronic medical records.

2. The method according to claim 1, characterized in that, The preset subjective-objective correlation mapping table specifically includes: The preset subjective-objective association mapping table is a key-value pair data table with the terminology of the standard symptoms as the key and the name of one or more clinical test items that cause the standard symptoms as the index.

3. The method according to claim 2, characterized in that, The step of determining whether the largest objective evidence support coefficient among all anchor pairs corresponding to the candidate symptom word is greater than a preset second threshold specifically includes: The preset subjective-objective association mapping table also stores a preset pathological association weight value for each key-value pair. The pathological association weight value represents the probability that an abnormality in the clinical test item corresponding to the index value will cause the standard symptom corresponding to the key value to occur. For each anchor pair, the objective evidence support coefficient corresponding to the anchor pair is multiplied by the preset pathological association weight value corresponding to the preset subjective-objective association mapping table to obtain the comprehensive pathological explanation index. Determine whether the maximum comprehensive pathological explanatory index among all the anchor pairs corresponding to the candidate symptom words is greater than a preset second threshold.

4. The method according to claim 3, characterized in that, The step of replacing the candidate symptom words with the standard medical terms corresponding to the associated indicator items in the preset subjective-objective association mapping table specifically includes: The preset subjective-objective correlation mapping table also stores a degree modifier classification table associated with the clinical test items. The degree modifier classification table defines the degree modifiers corresponding to the comprehensive pathological interpretation index in different numerical ranges. Based on the comprehensive pathological interpretation index of the target anchor pair corresponding to the maximum comprehensive pathological interpretation index, search for the matching target degree modifier in the degree modifier classification table; The target degree modifier and the target anchor pair are semantically concatenated with the standard medical terms corresponding to the preset subjective-objective association mapping table to generate a combined empirical semantic item. Replace the candidate symptom words with the combined empirical semantic terms.

5. The method according to claim 1, characterized in that, The step of calculating the objective evidence support coefficient based on the deviation of the detected value of the associated indicator item from the reference range interval of the associated indicator item specifically includes: When the detected value of the associated indicator is numerical, the reference range includes an upper threshold and a lower threshold. If the detected value of the associated indicator item is greater than the upper limit threshold, then the difference between the detected value and the upper limit threshold is calculated; The ratio of the difference to the upper limit threshold is determined as the objective evidence support coefficient; If the detected value of the associated indicator item is less than the lower limit threshold, then the difference between the lower limit threshold and the detected value is calculated. The ratio of the difference to the lower threshold is determined as the objective evidence support coefficient; If the detected value is between the upper threshold and the lower threshold, then the objective evidence support coefficient is determined to be zero.

6. The method according to claim 5, characterized in that, When the detected value of the associated indicator is numerical, after the step of including an upper threshold and a lower threshold in the reference range interval, the method further includes: When the detection value of the associated indicator item is qualitative text type, a preset qualitative quantization mapping table is obtained. The preset quantization mapping table stores the mapping relationship between different qualitative text type results and preset quantization coefficient values. Find the target preset quantization coefficient value that matches the detected value in the qualitative quantization mapping table; The target preset quantification coefficient value is determined as the objective evidence support coefficient.

7. The method according to claim 1, characterized in that, After the step of labeling the replaced candidate symptom words as empirical symptom data, the method further includes: Obtain the target item name and target detection value of the associated indicator item in the anchor pair on which the empirical symptom data is based; The target project name and the target detection value are encapsulated into evidence tracing tags according to a preset tracing format; The evidence tracing label is attached to the end of the empirical symptom data.

8. A generating apparatus, characterized in that, The generating device includes: one or more processors and a memory; the memory is coupled to the one or more processors, the memory is used to store computer program code, the computer program code including computer instructions, and the one or more processors call the computer instructions to cause the generating device to perform the method as described in any one of claims 1-7.

9. A computer program product containing instructions, characterized in that, When the computer program product is run on the production device, the production device performs the method as described in any one of claims 1-7.

10. A computer-readable storage medium comprising instructions, characterized in that, When the instructions are executed on the generating device, the generating device performs the method as described in any one of claims 1-7.

Images