Health management method and device based on auxiliary diagnosis, equipment and storage medium

Abstract

This invention provides a health management method, device, equipment, and storage medium based on assisted diagnosis. The method includes: standardizing patient medical record data, examination report data, and health monitoring data to construct a digital patient file; performing inference calculations through multiple intelligent analysis models, and adaptively weighting and fusing them according to the reliability coefficients and uncertainty measures of each model to generate a comprehensive probability ranking of multiple candidate results; generating assisted diagnostic results and recommended examination items based on this ranking; and calculating risk assessment indicators by combining the temporal characteristics of historical health data to generate personalized dynamic monitoring plans and health intervention suggestions. This invention improves output stability through multi-model fusion, avoiding the limitations of a single model; provides rich references for clinical diagnosis through multi-candidate ranking; and achieves dynamic risk assessment through temporal analysis, forming a complete closed loop from assisted diagnosis to health management, thus improving reliability and practicality.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a health management method, apparatus, device, and storage medium based on assisted diagnosis. Background Technology

[0002] With the rapid development of medical informatization and intelligent technologies, artificial intelligence has been widely applied in the fields of medical auxiliary diagnosis and health management. Modern medical systems have accumulated massive amounts of patient data, including various types of medical data such as electronic medical records, medical images, laboratory test reports, and continuous physiological monitoring data. This data not only provides information support for disease diagnosis but also provides a data foundation for patients' long-term health management and disease prevention.

[0003] Currently, existing assisted diagnostic and health management systems generally employ a single deep learning model or analytical algorithm to process medical data. These systems are typically trained and optimized for specific types of data or specific diagnostic tasks, achieving good results within their focused domain. However, when faced with multi-source, heterogeneous data in real-world clinical environments, the stability and consistency of output results are difficult to guarantee due to limitations in the model's structure, the limitations of the training data, and sensitivity to specific data noise. When the quality of the input data fluctuates, the data distribution shifts, or situations are not fully covered during model training, a single model is prone to producing unstable prediction results. This instability not only affects the reliability of assisted diagnosis but also directly restricts the development of subsequent health management plans based on diagnostic results, resulting in the inability to provide patients with stable and reliable dynamic monitoring and health intervention recommendations. Summary of the Invention

[0004] The main objective of this invention is to solve the technical problem of insufficient stability of output results caused by the use of a single analysis model in existing assisted diagnosis and health management systems; This invention provides a health management method based on assisted diagnosis, the method comprising: The collected patient medical records, examination reports, and health monitoring data are analyzed and standardized to construct digital patient records. Based on the patient's digital records, multiple intelligent analysis models are used to perform inference calculations. Adaptive weighted fusion is then performed based on the reliability coefficients of each intelligent analysis model on the calibration dataset and the uncertainty measure of its output to generate a comprehensive probability ranking of multiple candidate results. Based on the comprehensive probability ranking, auxiliary diagnostic results and recommended examination items are generated; Based on the time-series characteristics of the auxiliary diagnostic results and the patient's historical health data, health risk assessment indicators are calculated, and personalized dynamic monitoring plans and health intervention recommendations are generated according to the risk assessment indicators and the patient's individual characteristics.

[0005] The present invention also provides a health management device based on assisted diagnosis, the health management device based on assisted diagnosis comprising: The data processing module is used to parse and standardize the collected patient medical record data, examination report data, and health monitoring data to build digital patient records; The fusion diagnostic module is used to perform inference calculations through multiple intelligent analysis models based on the patient's digital records, and to perform adaptive weighted fusion based on the reliability coefficients of each intelligent analysis model on the calibration dataset and the uncertainty measure of the output, to generate a comprehensive probability ranking of multiple candidate results. The diagnostic output module is used to generate auxiliary diagnostic results and recommended examination items based on the comprehensive probability sorting. The health management module is used to calculate health risk assessment indicators based on the time-series characteristics of the auxiliary diagnostic results and the patient's historical health data, and to generate personalized dynamic monitoring plans and health intervention suggestions based on the risk assessment indicators and the patient's individual characteristics.

[0006] The present invention also provides a health management device based on assisted diagnosis, comprising: a memory and at least one processor, wherein the memory stores instructions, and the memory and the at least one processor are interconnected via a circuit; the at least one processor invokes the instructions in the memory to cause the health management device based on assisted diagnosis to perform the steps of the above-described health management method based on assisted diagnosis.

[0007] The present invention also provides a computer-readable storage medium storing instructions that, when executed on a computer, cause the computer to perform the steps of the above-described health management method based on assisted diagnosis.

[0008] The aforementioned health management method, device, equipment, and storage medium based on assisted diagnosis constructs digital patient records by standardizing patient medical record data, examination report data, and health monitoring data. It then uses multiple intelligent analysis models for inference calculations, adaptively weighting and fusing them based on the reliability coefficients and uncertainty measures of each model to generate a comprehensive probability ranking of multiple candidate results. Based on this ranking, it generates assisted diagnostic results and recommended examination items. Finally, it calculates risk assessment indicators by combining the temporal characteristics of historical health data, generating personalized dynamic monitoring plans and health intervention suggestions. This invention improves output stability through multi-model fusion, avoiding the limitations of a single model; provides rich references for clinical diagnosis through multi-candidate ranking; and achieves dynamic risk assessment through temporal analysis, forming a complete closed loop from assisted diagnosis to health management, thus improving reliability and practicality.

[0009] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention are realized and obtained in accordance with the structures particularly pointed out in the description, claims and drawings.

[0010] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0011] Figure 1 This is a schematic diagram of the first embodiment of the health management method based on assisted diagnosis in this invention. Figure 2 This is a schematic diagram of a second embodiment of the health management method based on assisted diagnosis in this invention. Figure 3 This is a schematic diagram of one embodiment of the health management device based on assisted diagnosis according to the present invention; Figure 4 This is a schematic diagram of one embodiment of a health management device based on assisted diagnosis in this invention. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0013] The terms "comprising" and "having," and any variations thereof, used in the embodiments of this invention are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the steps or units listed, but may optionally include other steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0014] To facilitate understanding of this embodiment, a health management method based on assisted diagnosis disclosed in this invention will first be described in detail. For example... Figure 1 As shown, this method includes the following steps: 101. Analyze and standardize the collected patient medical record data, examination report data, and health monitoring data to construct digital patient files; In this embodiment, the step of parsing and standardizing the collected patient medical record data, examination report data, and health monitoring data to construct a digital patient profile includes: parsing the patient medical record data, identifying key medical entities through a sequence labeling model and mapping them to a medical standard terminology library, performing unit conversion and interval normalization on the numerical indicators in the examination report data, and sorting and aligning the time series data in the health monitoring data according to timestamps to obtain a standardized medical information set; and integrating the standardized medical information set according to a preset data structure to generate a digital patient profile containing basic patient information, a symptom list, a list of test indicators, and historical monitoring sequences.

[0015] Specifically, because patient medical data typically originates from different medical systems and devices, it suffers from inconsistencies in format, terminology, and numerical units, necessitating systematic analysis and standardization. In terms of medical record data analysis, the system first performs word segmentation on the input medical record text, dividing the continuous text sequence into meaningful word units. This word segmentation differs from general Chinese word segmentation; it is optimized for the characteristics of medical text, enabling accurate identification of medical terminology and disease names. For example, "acute myocardial infarction" needs to be recognized as a complete medical concept, rather than being broken down into multiple independent words.

[0016] After word segmentation, the system identifies key medical entities in the text using a sequence labeling model. This model employs labeling methods based on conditional random fields or neural networks, capable of identifying different types of medical entities in the text, such as symptom descriptions, disease names, medication information, and examination items. This labeling process goes beyond simple keyword matching, considering the contextual relationships and semantic connections between words, thus handling situations where the same word expresses different medical concepts in different contexts. It's important to note that medical entity recognition also needs to handle negative and uncertain expressions; for example, the entity "diabetes" in "no history of diabetes" needs to be marked as negative to avoid misjudgment.

[0017] Identified medical entities need to be further mapped to a standard medical terminology database. In practical applications, the same medical concept may have multiple different expressions; for example, "hypertension" may be recorded as "high blood pressure," "elevated blood pressure," or the more specialized "primary hypertension." By mapping to a standard terminology database, these different expressions can be unified into standardized medical terminology codes, thereby making medical record data from different sources comparable. Commonly used standard terminology databases include the ICD coding system and SNOMED CT, among other internationally recognized medical terminology systems. The mapping process uses a semantic similarity calculation method. The system calculates the similarity between the identified medical entities and each term in the standard terminology database, selecting the term with the highest similarity exceeding a threshold as the mapping result.

[0018] The key to processing test report data lies in the standardization of numerical indicators. Different medical institutions and testing equipment may use different units of measurement; for example, blood glucose concentration may be measured in mmol / L or mg / dL. The system needs to convert all numerical indicators to a unified unit system according to preset unit conversion rules. Furthermore, since the numerical ranges of different test indicators vary greatly, directly using the raw values ​​can cause subsequent analysis models to become overly sensitive to indicators with large values. Therefore, interval normalization is necessary. Normalization methods typically employ maximum-minimum normalization or standardization methods to map indicators with different dimensions to the same numerical range. This processing does not change the distribution characteristics and relative relationships of the data, but it makes different indicators comparable.

[0019] The processing of health monitoring data primarily focuses on the alignment of time series data. Since monitoring data is continuously collected, different types of physiological parameters may have different sampling frequencies; for example, blood pressure may be measured hourly, while heart rate may be recorded every minute. To facilitate the analysis of the correlations and trends in these data, it is necessary to align the data according to a unified time benchmark. The system sorts the monitoring data according to timestamps and uses time windows or interpolation methods to align data with different frequencies to the same time point. For missing time points, reasonable interpolation can be used to supplement them based on preceding and following data points; no specific limitations are imposed on this.

[0020] After standardizing the data, the system integrates the processed information according to a pre-defined data structure. This data structure defines how the patient's digital records are organized, typically using a hierarchical design. The basic information layer contains static attributes such as the patient's identity, age, and gender; the symptom list records the patient's current and historical symptoms, with each symptom entry including attributes such as symptom name, onset time, and severity; the laboratory indicator list stores standardized test results, including indicator name, value, unit, and test time; and the historical monitoring sequence stores continuously monitored physiological parameter changes in chronological order.

[0021] 102. Based on the patient's digital records, multiple intelligent analysis models are used to perform reasoning calculations, and adaptive weighted fusion is performed based on the reliability coefficients of each intelligent analysis model on the calibration dataset and the uncertainty measure of its output to generate a comprehensive probability ranking of multiple candidate results. In this embodiment, the step of performing inference calculations using multiple intelligent analysis models based on the patient's digital records includes: inputting text data into a text inference model, numerical data into a statistical analysis model, and image data into a visual recognition model based on different data types in the patient's digital records to obtain initial feature representations for each model; each intelligent analysis model performing inference calculations based on the initial feature representations for a preset set of candidate categories to obtain a score value for each candidate category; and normalizing the score values ​​of each intelligent analysis model to generate a distribution vector containing each candidate category and its corresponding probability value.

[0022] Specifically, because patient digital records contain various types of medical data, each with its own characteristics and information expression methods, targeted analysis models are needed for processing. For textual data such as medical records and symptom descriptions, the system uses a text reasoning model. The model converts the input text sequence into word vector representations, encodes these word vectors through a multi-layer neural network, and extracts the deep semantic features of the text. These features not only include the meaning of individual words but also the relationships between words and the overall semantic structure.

[0023] For numerical data such as test indicators and physiological parameters, the system uses statistical analysis models for processing. Since numerical data is already quantified information, the focus of the analysis is on discovering statistical regularities and anomalies among the values. The statistical analysis model calculates the correlation strength between various indicators and different disease categories based on the distribution characteristics and interrelationships of the numerical data. This correlation strength calculation considers the combined effects and mutual influences of multiple indicators.

[0024] When patient records contain medical imaging data, the system invokes a visual recognition model for image analysis. This model employs a convolutional neural network architecture, progressively extracting image features through multiple convolutional operations. Shallow convolutional layers extract basic visual features such as edges and textures, while deeper convolutional layers extract more abstract semantic features, enabling the identification of pathological features and abnormal patterns in the image. It's important to note that the system can still function normally even if some patient records do not contain imaging data; in this case, only the outputs of the text inference model and the statistical analysis model are used for subsequent fusion.

[0025] After feature extraction, each intelligent analysis model performs inference calculations against a pre-defined set of candidate categories. The inference process involves inputting the extracted features into a classification layer, typically composed of a fully connected neural network, which maps the feature space to the candidate category space. Each intelligent analysis model independently scores each candidate category, with the score reflecting the model's confidence level in believing the patient belongs to that category. Because the scoring mechanisms and numerical ranges of different models may vary, the system normalizes the scores by converting them into a probability distribution using a softmax function, ensuring comparability of the outputs from all models.

[0026] After obtaining the probability distributions of each model, the system performs adaptive weighted fusion. This weighted fusion dynamically adjusts the weights based on the reliability and determinism of each model's output. The reliability coefficient reflects the model's performance on historical data, while the uncertainty measure is obtained by calculating the information entropy of the probability distribution. Information entropy describes the dispersion of the probability distribution; a smaller entropy value indicates greater certainty in the model's predictions when the probability distribution is relatively concentrated. The system assigns greater weights to models with high reliability and low uncertainty. This adaptive weighting mechanism fully utilizes the strengths of each model while suppressing the impact of unreliable predictions. Finally, the candidate categories are ranked according to their probability values ​​to obtain a comprehensive probability ranking of the multiple candidate results.

[0027] 103. Based on the comprehensive probability ranking, generate auxiliary diagnostic results and recommended examination items; In this embodiment, the generation of auxiliary diagnostic results is based on a dual screening mechanism of probability threshold and ranking position. The system sets a probability threshold, and only candidate categories with probability values ​​exceeding this threshold are included in the auxiliary diagnostic results. This threshold setting considers the balance between accuracy and coverage required in clinical applications, avoiding the inclusion of irrelevant categories with excessively low probabilities while ensuring that no clinically significant possibilities are overlooked. Simultaneously, the system limits the number of auxiliary diagnostic results, typically retaining the top-ranking candidate categories. This provides sufficient reference information for clinical decision-making without overburdening the judgment process with too many candidates.

[0028] It should be noted that for each candidate category included in the auxiliary diagnostic results, the system also attaches corresponding supporting evidence. This evidence comes from key features identified by various intelligent analysis models during the reasoning process, such as symptom descriptions identified by text reasoning models, abnormal indicators discovered by statistical analysis models, and image features detected by visual recognition models. By providing this supporting evidence, the auxiliary diagnostic results become interpretable, allowing doctors to understand why the system provides a particular candidate category and the basis for its probability assessment.

[0029] Based on the generated auxiliary diagnostic results, the system further generates recommended examination items. The logic for generating recommended examination items is based on a diagnostic decision tree and differential diagnostic needs. The system internally maintains a medical knowledge base that stores the relationships between different disease categories and related examination items. When multiple candidate categories exist in the auxiliary diagnostic results, the system analyzes the diagnostic distinguishing points of these candidate categories, that is, which examination items can effectively differentiate these candidate categories. Furthermore, the system prioritizes recommending examination items with higher differential diagnostic value, thus obtaining the maximum diagnostic information gain with the fewest examinations.

[0030] Furthermore, the generation of recommended tests also considers the patient's existing test data. The system compares the tests already completed in the patient's digital record to avoid recommending duplicate tests. For tests with outdated results or those requiring dynamic monitoring, the system determines whether retesting is necessary based on time intervals and clinical guidelines. In some cases, the system will also recommend targeted diagnostic tests based on the highest probability candidate category in the auxiliary diagnostic results; these tests are often the gold standard for diagnosing the disease.

[0031] 104. Based on the auxiliary diagnostic results and the temporal characteristics of the patient's historical health data, calculate health risk assessment indicators, and generate personalized dynamic monitoring plans and health intervention recommendations according to the risk assessment indicators and the patient's individual characteristics.

[0032] In this embodiment, the step of calculating health risk assessment indicators based on the time-series characteristics of the auxiliary diagnostic results and the patient's historical health data, and generating personalized dynamic monitoring plans and health intervention suggestions based on the risk assessment indicators and the patient's individual characteristics includes: extracting time series of key physiological indicators from the patient's historical health data; performing trend analysis and anomaly detection on the time series using a time-series analysis model; and calculating health risk assessment indicators by combining the candidate categories with the highest probability in the auxiliary diagnostic results; determining the key monitoring indicators and monitoring frequencies based on the health risk assessment indicators; calculating the warning thresholds for each monitoring indicator based on the patient's individual characteristics; and generating a dynamic monitoring plan; retrieving corresponding intervention measures from a knowledge base based on the health risk assessment indicators; and adapting and adjusting the intervention measures according to the patient's individual characteristics to generate health intervention suggestions.

[0033] Specifically, the core of health risk assessment lies in identifying potential risk signals and evolving trends from a patient's historical health data. Unlike static analysis of single examination results, time-series feature analysis focuses on the patterns of change in physiological indicators over time. The system first extracts historical records of key physiological indicators from the patient's digital file. These indicators may include routine test results such as blood pressure, blood glucose, and blood lipids, or they may include continuous monitoring data from wearable devices. Each indicator is arranged chronologically to form a time series. Each data point in the time series not only records the indicator value but also includes the collection time and related contextual information.

[0034] Time series analysis models conduct in-depth analysis of these time series, focusing primarily on two characteristics. The first is trend characteristics, namely the overall direction and speed of change of indicator values ​​over a longer time span. For example, does an indicator show a continuous upward or downward trend? This trend reflects the evolution of the patient's health status. The second is abnormal fluctuation characteristics, namely, drastic changes in indicator values ​​within a short period or deviations from the normal range. Such abnormalities are often early signals of disease onset or changes in condition. Time series analysis models typically combine statistical analysis and machine learning techniques, enabling them to capture both obvious linear trends and identify complex nonlinear patterns.

[0035] It's important to note that the results of time-series feature analysis need to be combined with auxiliary diagnostic results to obtain a meaningful risk assessment. The system extracts the candidate categories with the highest probabilities from the auxiliary diagnostic results and uses them as a reference background for risk assessment. Different disease categories focus on different risk dimensions; for example, some diseases require a focus on complication risk, while others require a focus on disease progression or recurrence risk. Based on the characteristics of the candidate categories and combined with the trends and anomalies found in the time-series analysis, the system calculates multidimensional health risk assessment indicators for that category. These risk indicators quantitatively describe the probability of a patient experiencing adverse health events in the future.

[0036] After obtaining the health risk assessment indicators, the system generates a dynamic monitoring plan. The development of this plan follows a risk-oriented principle; for aspects with higher risk assessment indicators, the system will arrange more frequent monitoring and stricter surveillance. Generating the monitoring plan first requires identifying the monitoring indicators. Based on the risk assessment results, the system selects those indicators most sensitive to changes in risk as key monitoring targets. This then determines the monitoring frequency; the monitoring frequency for high-risk items will be increased accordingly to facilitate timely detection and intervention.

[0037] The warning threshold settings in the monitoring scheme reflect a personalized approach. Traditional warning thresholds often use uniform standard values, but this "one-size-fits-all" approach cannot adapt to the individual differences among patients. In this embodiment, the warning threshold calculation considers the patient's individual characteristics, including age, gender, medical history, and medication use. The system retrieves baseline warning thresholds for specific candidate categories from a knowledge base, and then calculates adjustment coefficients based on the patient's individual characteristics to correct the baseline thresholds. For example, elderly patients and younger patients have different tolerances to certain abnormal indicators, and their warning threshold settings should also differ. Through this personalized adjustment, the warning threshold can more accurately reflect the patient's actual risk threshold.

[0038] The generation of health intervention recommendations is also based on risk assessment results and individual characteristics. The system identifies high-risk items requiring intervention based on health risk assessment indicators and retrieves corresponding intervention measures from the medical knowledge base. Intervention measures may involve multiple aspects, such as lifestyle modifications, medication adjustments, and follow-up recommendations. The retrieved intervention measures are general recommendations based on clinical guidelines and evidence-based medicine, but in practical application, they need to be adapted to the individual patient's situation. The system analyzes the patient's adherence history, lifestyle habits, economic conditions, and other individual characteristics to adjust the expression, implementation difficulty, and intensity of the intervention measures, ensuring that the intervention recommendations are both clinically reasonable and practically feasible.

[0039] In this embodiment, a digital patient profile is constructed by standardizing patient medical record data, examination report data, and health monitoring data. Multiple intelligent analysis models are used for inference calculations, and adaptive weighted fusion is performed based on the reliability coefficients and uncertainty measures of each model to generate a comprehensive probability ranking of multiple candidate results. Based on this ranking, auxiliary diagnostic results and recommended examination items are generated. Risk assessment indicators are calculated by combining the temporal characteristics of historical health data to generate personalized dynamic monitoring plans and health intervention suggestions. This invention improves output stability through multi-model fusion, avoiding the limitations of a single model; provides rich references for clinical diagnosis through multi-candidate ranking; and achieves dynamic risk assessment through temporal analysis, forming a complete closed loop from auxiliary diagnosis to health management, thus improving reliability and practicality.

[0040] Please see Figure 2 Another embodiment of the health management method based on assisted diagnosis in this application includes: 201. Analyze and standardize the collected patient medical record data, examination report data, and health monitoring data to construct digital patient files; 202. Based on the patient's digital records, inference calculations are performed using multiple intelligent analysis models. 203. Obtain the reliability coefficients of each intelligent analysis model pre-calculated on the calibration dataset, and calculate the deterministic weights of each intelligent analysis model based on the distribution vectors of the current inference calculation output of each intelligent analysis model; In this embodiment, the step of calculating the deterministic weights of each intelligent analysis model based on the distribution vector currently output by each intelligent analysis model includes: calculating the information entropy of the probability values ​​of each candidate category in the distribution vector currently output by each intelligent analysis model to obtain the information entropy value of each intelligent analysis model; dividing the information entropy value by the theoretical maximum entropy value and performing normalization processing to obtain the uncertainty measure of each intelligent analysis model; and subtracting the uncertainty measure from 1 to obtain the deterministic weights of each intelligent analysis model.

[0041] Specifically, the distribution vector output by the intelligent analysis model after completing inference calculations reflects the model's judgment results for each candidate category, but different distribution patterns convey different levels of predictive confidence. When the probability value of a certain candidate category in a model's output distribution is significantly higher than that of other categories, it indicates that the model is relatively certain about its prediction results; conversely, when the probability values ​​of each candidate category in the output distribution are relatively close, it indicates that the model is hesitant among multiple candidates, and the certainty of the prediction is low. This difference in the degree of certainty needs to be taken into account during model fusion; otherwise, unreliable predictions will have an excessive impact on the final result.

[0042] Information entropy is a classic metric used to quantify the uncertainty of a probability distribution, and its calculation is based on information theory in probability theory. For a probability distribution containing multiple candidate categories, information entropy measures the degree of disorder or uncertainty of the distribution by performing a logarithmic operation on the probability values ​​of each category and summing the results. When the probability distribution is highly concentrated in one category, the probability values ​​of other categories are very small, resulting in a lower information entropy value, indicating a higher degree of certainty in the distribution. When the probability distribution is relatively uniform across multiple categories, the resulting information entropy value is higher, indicating greater uncertainty in the distribution. It should be noted that categories with a probability value of zero are not included in the logarithmic operation during the calculation of information entropy, because the amount of information corresponding to zero probability is theoretically infinite, and needs to be processed in actual calculations.

[0043] After obtaining the information entropy values ​​of the output distribution of each intelligent analysis model, the system needs to normalize these entropy values ​​to ensure fair comparison under different numbers of candidate categories. The theoretical maximum entropy value corresponds to the case of a perfectly uniform distribution, where all candidate categories have the same probability value, and the model's prediction uncertainty reaches its maximum in this case. By dividing the actual calculated information entropy value by the theoretical maximum entropy value, a normalized uncertainty measure between zero and one can be obtained. A value of zero indicates that the model's prediction is completely certain, and a value of one indicates that the model's prediction is completely uncertain.

[0044] The deterministic weights are calculated using a simple and intuitive conversion method: subtracting the normalized uncertainty measure from one. After this processing, the deterministic weights are still values ​​between zero and one, but their meaning is the opposite of the uncertainty measure; a larger weight value indicates a more certain prediction from the model. The rationale behind this conversion method is that, during model fusion, models with greater confidence in their predictions should be given larger weights, while models with uncertain predictions should be assigned smaller weights.

[0045] In practical applications, deterministic weights and reliability coefficients are used together to determine the contribution of each model in the fusion process. The reliability coefficient is an inherent property of the model, reflecting its overall performance on historical data; it is a static quality assessment indicator. Deterministic weights, on the other hand, are dynamically calculated based on the current input data, reflecting the model's confidence in predicting the current specific case. Combining the two considers both the model's long-term reliability and its applicability to the current case. A model with historically good performance will have its weight appropriately reduced if its prediction for the current case has high uncertainty; conversely, a model with mediocre historical performance will have its weight increased if it provides a highly certain prediction for the current case.

[0046] Furthermore, before obtaining the reliability coefficients pre-calculated by each intelligent analysis model on the calibration dataset, the method further includes: preparing a calibration dataset containing labeled data; inputting samples from the calibration dataset into each intelligent analysis model to obtain the prediction results of each intelligent analysis model for each sample; extracting the highest probability category from the prediction results of each intelligent analysis model and comparing it with the reference label of the corresponding sample to count the number of matching samples for each intelligent analysis model; dividing the number of matching samples for each intelligent analysis model by the total number of samples in the calibration dataset to obtain the reliability coefficients of each intelligent analysis model and saving them.

[0047] Specifically, the reliability coefficient is calculated during the preparation phase before the system's actual application. This process is essentially a performance evaluation and quality calibration of each intelligent analysis model. Unlike the training process, the reliability coefficient calculation does not involve adjusting model parameters. Instead, it obtains an objective measure of model quality by observing the model's performance in standardized test scenarios. This measurement result is saved and used as a weighting basis for model fusion in subsequent practical applications.

[0048] Preparing the calibration dataset is fundamental to the entire evaluation process. This dataset needs to meet several key characteristics. First, the data must contain accurate annotation information; each sample must have a clear reference label, typically provided by professionals based on clinical practice, ensuring high reliability. Second, the data distribution should match the actual application scenario. If the distribution of the calibration dataset differs significantly from the actual data, the reliability coefficient obtained will fail to accurately reflect the model's performance in real-world applications. Third, the dataset size needs to meet statistical requirements; too small a sample size will lead to significant susceptibility to random factors in the evaluation results, making them unstable and unreliable.

[0049] After preparing the calibration dataset, the system sequentially inputs each sample from the dataset into each intelligent analysis model. This process is completely consistent with the inference process in actual applications, ensuring consistency between the evaluation environment and the application environment. Each model performs inference calculations based on the input samples and outputs a distribution vector containing the probability values ​​of each candidate category. It should be noted that at this stage, the models run independently, there is no information exchange between the models, and each model gives a prediction result based on its own judgment.

[0050] For each model's output probability distribution vector, the system extracts the class with the highest probability value as the model's predicted class for the current sample. This approach reflects the model's tendency to make a single choice when forced to do so. Although the model outputs a complete probability distribution, accuracy evaluation typically focuses on whether the class the model considers most likely is correct. Extracting the highest probability class is technically simple: it involves traversing the probability vector to find the element with the largest value and its corresponding class label.

[0051] Comparing the predicted category with the sample's reference label is a crucial step in determining the accuracy of the prediction. The comparison process employs a strict matching principle: a prediction is considered correct only when the predicted category perfectly matches the reference label; any deviation is considered a prediction error. While this stringent standard may underestimate the model's actual capabilities, it ensures that the reliability coefficient reflects the model's ability to make completely accurate judgments, a critical safety consideration in medical applications. The system performs this comparison on every sample in the calibration dataset and counts the number of samples correctly predicted by each model.

[0052] The reliability coefficient of a model is the ratio of the number of matched samples to the total number of samples. This ratio essentially represents the model's accuracy on the calibration dataset. The reliability coefficient is a value between zero and one; the closer the value is to one, the higher the model's reliability, and the closer it is to zero, the lower the model's reliability. Ideally, a perfect model should have a reliability coefficient of one, meaning that its predictions for all samples are correct. A model that relies entirely on random guessing will have a reliability coefficient close to the reciprocal of the number of candidate classes. In practice, the reliability coefficient of intelligent analysis models typically falls between these two extremes, with the specific value depending on the model's design quality, training effectiveness, and the complexity of the task.

[0053] The calculated reliability coefficients are saved by the system and associated with the corresponding models. In subsequent practical applications, whenever multi-model fusion is required, the system directly reads these pre-calculated reliability coefficients without needing to recalculate them. This design ensures both the efficiency of the fusion process and the standardization and repeatability of reliability assessment. If the model is retrained or updated, the reliability coefficients need to be recalculated on the calibration dataset to ensure that the coefficient values ​​accurately reflect the current state of the model.

[0054] 204. For each candidate category, multiply the probability value of each intelligent analysis model for the corresponding candidate category by the reliability coefficient and deterministic weight of the corresponding model, and then sum them to obtain the fusion score of each candidate category. In this embodiment, the fusion score is calculated using a weighted summation method. For any candidate category, each intelligent analysis model outputs a probability value, but the reliability of these probability values ​​varies. The reliability coefficient is derived from the model's historical performance on the calibration dataset, reflecting the overall quality level of the model. The deterministic weight is calculated based on the distribution entropy value currently output by the model, reflecting the model's confidence in predicting the current case.

[0055] In the specific calculation process, the system multiplies the probability value output by the model by its reliability coefficient, and then by the deterministic weights to obtain the model's weighted contribution to the candidate category. This dual-weighting design gives higher weights to models with good historical performance and certain current predictions, while reducing the weights of models with poor historical performance or hesitant current predictions. Summing the weighted contributions of all models yields the fusion score for the candidate category. This process is performed separately for each candidate category, ultimately resulting in a sequence of fusion scores.

[0056] 205. Normalize the fusion scores of all candidate categories and sort them from high to low according to the normalized values ​​to generate a comprehensive probability ranking of multiple candidate results. In this embodiment, the fusion scores, after weighted summation, need to be converted into probabilistic form to be reasonably interpreted. The normalization method is to divide the fusion score of each candidate category by the sum of the fusion scores of all candidate categories. After this processing, the sum of the values ​​of each candidate category equals one, and each value can be understood as the comprehensive probability of that candidate category.

[0057] After normalization, the system sorts the candidate categories in descending order based on their probability values. The candidate category with the highest probability value is ranked first, representing the most likely outcome as determined by the system; other candidate categories are then arranged in descending order of probability. This sorting preserves all candidate categories and their probability information, providing both the primary judgment and other possibilities, thus meeting the practical needs of clinical differential diagnosis.

[0058] 206. Based on the comprehensive probability ranking, generate auxiliary diagnostic results and recommended examination items; 207. Based on the auxiliary diagnostic results and the temporal characteristics of the patient's historical health data, calculate health risk assessment indicators, and generate personalized dynamic monitoring plans and health intervention recommendations according to the risk assessment indicators and the patient's individual characteristics.

[0059] In this embodiment, a digital patient profile is constructed by standardizing patient medical record data, examination report data, and health monitoring data. Multiple intelligent analysis models are used for inference calculations, and adaptive weighted fusion is performed based on the reliability coefficients and uncertainty measures of each model to generate a comprehensive probability ranking of multiple candidate results. Based on this ranking, auxiliary diagnostic results and recommended examination items are generated. Risk assessment indicators are calculated by combining the temporal characteristics of historical health data to generate personalized dynamic monitoring plans and health intervention suggestions. This invention improves output stability through multi-model fusion, avoiding the limitations of a single model; provides rich references for clinical diagnosis through multi-candidate ranking; and achieves dynamic risk assessment through temporal analysis, forming a complete closed loop from auxiliary diagnosis to health management, thus improving reliability and practicality.

[0060] The above describes the health management method based on assisted diagnosis in the embodiments of the present invention. The following describes the health management device based on assisted diagnosis in the embodiments of the present invention. Please refer to [link to relevant documentation] for details on this assisted diagnosis health management device. Figure 3 One embodiment of the health management device based on assisted diagnosis in this invention includes: Data processing module 301 is used to parse and standardize the collected patient medical record data, examination report data and health monitoring data to build digital patient archives; The fusion diagnostic module 302 is used to perform inference calculations through multiple intelligent analysis models based on the patient's digital records, and to perform adaptive weighted fusion based on the reliability coefficients of each intelligent analysis model on the calibration dataset and the uncertainty measure of the output, to generate a comprehensive probability ranking of multiple candidate results. The diagnostic output module 303 is used to generate auxiliary diagnostic results and recommended examination items based on the comprehensive probability sorting. The health management module 304 is used to calculate health risk assessment indicators based on the time-series characteristics of the auxiliary diagnostic results and the patient's historical health data, and to generate personalized dynamic monitoring plans and health intervention suggestions based on the risk assessment indicators and the patient's individual characteristics.

[0061] In this embodiment of the invention, the health management device based on assisted diagnosis operates the aforementioned health management method based on assisted diagnosis. The device standardizes patient medical record data, examination report data, and health monitoring data to construct a digital patient file; it performs inference calculations using multiple intelligent analysis models, and adaptively weights and fuses these models based on their reliability coefficients and uncertainty metrics to generate a comprehensive probability ranking of multiple candidate results; based on this ranking, it generates assisted diagnostic results and recommended examination items; and it calculates risk assessment indicators by combining the temporal characteristics of historical health data to generate personalized dynamic monitoring plans and health intervention suggestions. This invention improves output stability through multi-model fusion, avoiding the limitations of a single model; it provides rich references for clinical diagnosis through multi-candidate ranking; and it achieves dynamic risk assessment through temporal analysis, forming a complete closed loop from assisted diagnosis to health management, thus improving reliability and practicality.

[0062] above Figure 3 The health management device based on assisted diagnosis in the embodiments of the present invention will be described in detail from the perspective of unitized functional entities. The health management device based on assisted diagnosis in the embodiments of the present invention will be described in detail from the perspective of hardware processing.

[0063] Figure 4 This is a schematic diagram of the structure of a health management device based on assisted diagnosis provided in an embodiment of the present invention. The health management device 400 based on assisted diagnosis can vary considerably due to differences in configuration or performance. It may include one or more central processing units (CPUs) 410 (e.g., one or more processors) and a memory 420, and one or more storage media 430 (e.g., one or more mass storage devices) for storing application programs 433 or data 432. The memory 420 and storage media 430 can be temporary or persistent storage. The program stored in the storage media 430 may include one or more units (not shown in the diagram), each unit may include a series of instruction operations on the health management device 400 based on assisted diagnosis. Furthermore, the processor 410 may be configured to communicate with the storage media 430 and execute the series of instruction operations in the storage media 430 on the health management device 400 based on assisted diagnosis to implement the steps of the aforementioned health management method based on assisted diagnosis.

[0064] The health management device 400 based on assisted diagnosis may also include one or more power supplies 440, one or more wired or wireless network interfaces 450, one or more input / output interfaces 460, and / or one or more operating systems 431, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 4 The illustrated structure of the health management device based on assisted diagnosis does not constitute a limitation on the health management device based on assisted diagnosis provided by the present invention. It may include more or fewer components than illustrated, or combine certain components, or have different component arrangements.

[0065] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the health management method based on assisted diagnosis.

[0066] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the system, device, or unit described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0067] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention 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 spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A health management method based on assisted diagnosis, characterized in that, The health management method based on assisted diagnosis includes: The collected patient medical records, examination reports, and health monitoring data are analyzed and standardized to construct digital patient records. Based on the patient's digital records, multiple intelligent analysis models are used to perform inference calculations. Adaptive weighted fusion is then performed based on the reliability coefficients of each intelligent analysis model on the calibration dataset and the uncertainty measure of its output to generate a comprehensive probability ranking of multiple candidate results. Based on the comprehensive probability ranking, auxiliary diagnostic results and recommended examination items are generated; Based on the time-series characteristics of the auxiliary diagnostic results and the patient's historical health data, health risk assessment indicators are calculated, and personalized dynamic monitoring plans and health intervention recommendations are generated according to the risk assessment indicators and the patient's individual characteristics.

2. The health management method based on assisted diagnosis according to claim 1, characterized in that, The process of parsing and standardizing the collected patient medical record data, examination report data, and health monitoring data to construct digital patient records includes: The patient medical record data is parsed, key medical entities are identified through a sequence labeling model and mapped to a medical standard terminology database, unit conversion and interval normalization are performed on numerical indicators in the examination report data, and time series data in the health monitoring data are sorted and aligned according to timestamps to obtain a standardized medical information set. The standardized medical information set is integrated according to a preset data structure to generate a digital patient file containing basic patient information, a list of symptoms, a list of test indicators, and historical monitoring sequences.

3. The health management method based on assisted diagnosis according to claim 1, characterized in that, The step of performing reasoning calculations based on the patient's digital records using multiple intelligent analysis models includes: Based on the different data types in the patient's digital records, text data is input into a text reasoning model, numerical data is input into a statistical analysis model, and image data is input into a visual recognition model to obtain the initial feature representations of each model. Each intelligent analysis model performs inference calculations based on the initial feature representation and a preset set of candidate categories to obtain the score value of each model for each candidate category; The scores of each intelligent analysis model are normalized to generate a distribution vector containing each candidate category and its corresponding probability value.

4. The health management method based on assisted diagnosis according to claim 1, characterized in that, The adaptive weighted fusion based on the reliability coefficients and output uncertainty measures of each intelligent analysis model on the calibration dataset to generate a comprehensive probability ranking of multiple candidate results includes: Obtain the reliability coefficients of each intelligent analysis model pre-calculated on the calibration dataset, and calculate the deterministic weights of each intelligent analysis model based on the current output distribution vector of each intelligent analysis model; For each candidate category, the probability value of each intelligent analysis model for the corresponding candidate category is multiplied by the reliability coefficient and deterministic weight of the corresponding model, and then summed to obtain the fusion score of each candidate category; The fusion scores of all candidate categories are normalized and sorted from high to low according to the normalized values ​​to generate a comprehensive probability ranking of multiple candidate results.

5. The health management method based on assisted diagnosis according to claim 4, characterized in that, The calculation of the deterministic weights of each intelligent analysis model based on the distribution vector currently output by each intelligent analysis model includes: The information entropy of each candidate category in the current output distribution vector of each intelligent analysis model is calculated to obtain the information entropy value of each intelligent analysis model. Divide the information entropy value by the theoretical maximum entropy value and normalize it to obtain the uncertainty measure of each intelligent analysis model; Subtracting the uncertainty measure from 1 yields the deterministic weights of each intelligent analysis model.

6. The health management method based on assisted diagnosis according to claim 4, characterized in that, Before obtaining the reliability coefficients of each intelligent analysis model pre-calculated on the calibration dataset, the method further includes: Prepare a calibration dataset containing labeled data, input the samples in the calibration dataset into each intelligent analysis model, and obtain the prediction results of each intelligent analysis model for each sample; Extract the highest probability category from the prediction results of each intelligent analysis model and compare it with the reference label of the corresponding sample to count the number of matching samples for each intelligent analysis model. Divide the number of matching samples for each intelligent analysis model by the total number of samples in the calibration dataset to obtain the reliability coefficient of each intelligent analysis model and save it.

7. The health management method based on assisted diagnosis according to claim 1, characterized in that, The calculation of health risk assessment indicators based on the time-series characteristics of the auxiliary diagnostic results and the patient's historical health data, and the generation of personalized dynamic monitoring plans and health intervention recommendations based on the risk assessment indicators and the patient's individual characteristics, include: The time series of key physiological indicators are extracted from the patient's historical health data. The time series are then analyzed for trends and anomalies using a time series analysis model. Combined with the candidate category with the highest probability in the auxiliary diagnostic results, a health risk assessment index is calculated. Based on the aforementioned health risk assessment indicators, the key indicators and monitoring frequencies that need to be monitored are determined, and the warning thresholds for each monitoring indicator are calculated based on the individual characteristics of the patients, thereby generating a dynamic monitoring plan. Based on the health risk assessment indicators, corresponding intervention measures are retrieved from the knowledge base, and the intervention measures are adapted and adjusted according to the individual characteristics of the patient to generate health intervention recommendations.

8. A health management device based on assisted diagnosis, characterized in that, The health management device based on assisted diagnosis includes: The data processing module is used to parse and standardize the collected patient medical record data, examination report data, and health monitoring data to build digital patient records; The fusion diagnostic module is used to perform inference calculations through multiple intelligent analysis models based on the patient's digital records, and to perform adaptive weighted fusion based on the reliability coefficients of each intelligent analysis model on the calibration dataset and the uncertainty measure of the output, to generate a comprehensive probability ranking of multiple candidate results. The diagnostic output module is used to generate auxiliary diagnostic results and recommended examination items based on the comprehensive probability sorting. The health management module is used to calculate health risk assessment indicators based on the time-series characteristics of the auxiliary diagnostic results and the patient's historical health data, and to generate personalized dynamic monitoring plans and health intervention suggestions based on the risk assessment indicators and the patient's individual characteristics.

9. A health management device based on assisted diagnosis, characterized in that, The health management device based on assisted diagnosis includes: a memory and at least one processor, wherein the memory stores instructions; The at least one processor invokes the instructions in the memory to cause the assisted diagnostics-based health management device to perform the steps of the assisted diagnostics-based health management method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the steps of the health management method based on assisted diagnosis as described in any one of claims 1-7.

Images