Adaptive disease clinical diagnosis method, device, equipment and medium

Abstract

This invention discloses an adaptive clinical diagnostic method, device, equipment, and medium for diseases, relating to the interdisciplinary fields of artificial intelligence and medical informatics. It achieves multidimensional knowledge distillation and adaptive diagnosis: this method not only aligns with the final prediction distribution of the teacher model but also jointly distills its inference chain and intermediate representations, enabling the student model to possess both diagnostic and reasoning capabilities. Based on this, an adaptive weighting mechanism based on teacher uncertainty dynamically adjusts the learning and diagnostic paths according to case complexity, thereby significantly improving diagnostic accuracy and training stability.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of artificial intelligence and medical informatics, and in particular to an adaptive clinical diagnosis method, apparatus, equipment, and medium for diseases based on clinical guideline structural constraints and data-driven multidimensional distillation. Background Technology

[0002] Liver disease, a serious global health issue, poses a significant challenge to human health due to its high incidence and mortality rates. Liver diseases are diverse, with complex and variable disease progressions exhibiting significant heterogeneity and dynamic changes, making the timing of diagnosis and treatment a major challenge. The complex clinical manifestations of liver disease require physicians to possess profound professional knowledge and extensive practical experience. In reality, high-level liver disease specialists are relatively scarce and unevenly distributed, making it difficult for many regions to obtain timely and accurate diagnostic services.

[0003] Furthermore, traditional diagnostic methods often rely on the experience and judgment of individual physicians, making them susceptible to subjective influences and lacking standardization and reproducibility. Therefore, there is an urgent need to establish an objective and reliable diagnostic model for liver diseases to improve the accuracy of comprehensive diagnosis of complex liver conditions. Given the seriousness of medical diagnosis, the most important aspect of this method is its transparency and interpretability, enabling healthcare professionals to understand and trust the diagnostic process and results, thus facilitating its application in clinical practice.

[0004] Therefore, the application of intelligent liver disease diagnostic models in clinical liver disease analysis must meet the following conditions: First, a standardized diagnostic pathway. The model should follow clinical guidelines and best practices, constructing a standardized diagnostic pathway to ensure that each step of the diagnosis has clear evidence and procedures, thereby improving the consistency and reliability of the diagnosis, while facilitating learning and application by medical personnel. Second, adaptive disease management capability. Faced with the diversity and complexity of liver diseases, the model needs to have adaptive disease management capability, that is, to be able to dynamically adjust the diagnostic strategy according to the patient's specific situation, regardless of the severity or urgency of the disease or the different accompanying symptoms, and to make corresponding diagnostic responses, improving the flexibility and specificity of the diagnosis. Finally, comprehensive analysis and interpretability. To improve the credibility of diagnostic results, the model must be able to analyze all possible types of liver diseases to the greatest extent and provide clear diagnostic evidence, including but not limited to etiological analysis, explanation of pathophysiological mechanisms, and reasons for treatment recommendations, ensuring that every diagnostic conclusion is supported by sufficient evidence and can be presented to medical professionals in an easily understandable way, making it highly interpretable and practical in actual clinical work.

[0005] However, existing liver disease diagnostic models mainly rely on single models or empirical ensemble methods, making it difficult to take into account individual patient differences and case complexity. They also lack comprehensive multi-faceted analysis of complex or uncertain cases. At the same time, existing distillation strategies usually only align the final output distribution, ignoring the transmission of the teacher's reasoning process and intermediate representations. Furthermore, they lack structured constraints that explicitly introduce clinical guidelines and knowledge graphs into the training process. This makes it difficult for student models to maintain accuracy while ensuring consistency of clinical reasoning logic and training stability, thus limiting their practicality and reliability in low-computing-power, edge deployment scenarios. Summary of the Invention

[0006] In view of the above problems, this invention provides an adaptive clinical diagnostic method, apparatus, device, and medium for overcoming or at least partially solving the above problems. It realizes the answer distribution, inference chain, and intermediate representation of a joint distillation teacher model, and constructs logical consistency constraints by combining a clinical knowledge graph. Simultaneously, it completes dynamic learning that matches case complexity through an adaptive sample weighting mechanism based on uncertainty. This method not only matches the optimal diagnostic path, significantly improving the accuracy and efficiency of diagnosis, but also enhances the transparency and credibility of the model through structured logic, improving the reliability of the diagnostic process and the model's generalization ability. Especially when dealing with complex or uncertain cases, it exhibits stable and interpretable diagnostic results.

[0007] This invention provides the following solution:

[0008] An adaptive clinical diagnostic method for diseases, comprising: The process involves acquiring the patient's electronic medical record dataset, dividing and standardizing the information contained in the electronic medical record dataset, and establishing a multimodal feature vector. The electronic medical record dataset includes at least medical history information, imaging examinations, biochemical tests, and physical examination records. We analyze disease diagnosis and treatment guidelines and medical knowledge graphs, transform the diagnosis and treatment logic into a digital reasoning structure, extract key clinical features and causal links, and embed them into student model training in the form of soft logic constraints. The system introduces answer distribution, reasoning chain, and intermediate representation information provided by the teacher model, and guides the student model to learn a unified representation of diagnostic and reasoning abilities through multi-dimensional distillation. Based on the uncertainty of the prediction results of the teacher model, an adaptive weighting mechanism is generated. This adaptive weighting mechanism is used to dynamically evaluate and stratify the complexity of cases, dividing cases into different complexity levels: simple, medium, and complex. The student model is adaptively invoked based on the complexity of different cases for reasoning and judgment. For simple cases, the student model directly provides a rapid diagnosis. For moderate cases, a structured logic constraint reinforcement model is introduced to constrain the clinical pathway. For complex cases, a multi-level comprehensive analysis is performed by combining the reasoning chain of multi-dimensional distillation with knowledge graph logic verification to ensure that the diagnostic conclusion achieves a balance between accuracy, logical consistency and interpretability.

[0009] Preferably: the disease diagnosis and treatment guidelines are analyzed to extract typical rules; fuzzy logic is used to quantify the overall degree of rule satisfaction; During training, if the predictions given by the student model conflict with clinical rules, the following penalty function is used for punishment:

[0010] In the formula: Indicates sample In the rules Below are the prediction results The rule satisfaction function.

[0011] Preferably, the multidimensional distillation includes answer distillation, reasoning distillation, and characterization distillation.

[0012] Preferably: the answer distillation includes using KL divergence to make the prediction distribution of the student model approximate that of the teacher model; the inference distillation includes learning the results of the teacher model and learning the inference process of the teacher model, including inference chains or attention distributions; the representation distillation includes aligning intermediate layer feature representations so that the student model learns the internal representations of the teacher model.

[0013] Preferably, the adaptive weighting mechanism includes a complexity scoring function and an adaptive weight calculation function, wherein the complexity scoring function is expressed by the following formula:

[0014] In the formula: express, express; The adaptive weight calculation function is expressed by the following formula:

[0015] In the formula: This represents the weighting adjustment coefficient. Indicates the output distribution of the teacher model entropy, The teacher model represents the category. The predicted probability.

[0016] Preferably, the final diagnosis is generated through a joint scoring mechanism of probability distribution and rule satisfaction, and a clear evidentiary interpretation is provided; The student model outputs a probability distribution for each disease category and a candidate inference chain based on the input electronic medical record features. Based on medical knowledge graphs and clinical guidelines, calculate the degree to which this case meets different diagnostic rules; The student model's predicted probability is combined with the rule satisfaction level to obtain a comprehensive scoring function; After obtaining the comprehensive scores of all candidate diseases, the disease with the highest score is selected as the final diagnosis result; Along with the diagnostic conclusion, a diagnostic evidence report is generated, which includes a positive evidence set and a missing or insufficient evidence set.

[0017] Preferably, the comprehensive scoring function is expressed by the following formula:

[0018] In the formula: This indicates that the student model is based on the input samples. The following prediction results are The logarithmic probability, Indicates the rule weight coefficient. Represents the sample complexity function. The logarithm of the rule consistency term represents the prediction result. In the sample The degree of alignment with clinical rules, This indicates the degree of rule satisfaction.

[0019] A clinical diagnostic device for diseases, used to perform the adaptive clinical diagnostic method for diseases described above, the device comprising: The electronic medical record data input and preprocessing unit is used to acquire the patient's electronic medical record dataset, divide and standardize the information contained in the electronic medical record dataset, and establish a multimodal feature vector; the electronic medical record dataset includes at least medical history information, imaging examinations, biochemical tests, and physical examination records; The clinical knowledge structure modeling unit is used to analyze disease diagnosis and treatment guidelines and medical knowledge graphs, transform the diagnosis and treatment logic into digital reasoning structures, extract key clinical features and causal links, and embed them into student model training in the form of soft logic constraints. A multidimensional knowledge distillation and adaptive classification unit is used to introduce the answer distribution, reasoning chain and intermediate representation information provided by the teacher model. Through multidimensional distillation, the student model is guided to learn a unified representation of diagnostic ability and reasoning ability. An adaptive weighting mechanism is generated based on the uncertainty of the prediction results of the teacher model. The adaptive weighting mechanism is used to dynamically evaluate and stratify the case complexity, classifying cases into different complexity levels of simple, medium and complex. The collaborative analysis and comprehensive diagnosis unit is used to adaptively call the distilled student model according to the complexity of different cases for reasoning and judgment. For simple cases, the student model is used to directly provide a rapid diagnosis. For moderate cases, a structured logic constraint reinforcement model is introduced to constrain the clinical pathway. For complex cases, a multi-level comprehensive analysis is performed by combining the reasoning chain of multi-dimensional distillation and the logical verification of knowledge graph to ensure that the diagnostic conclusion achieves a balance between accuracy, logical consistency and interpretability.

[0020] A clinical diagnostic device for a disease, the device comprising a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the above-described adaptive clinical diagnosis method for diseases according to the instructions in the program code.

[0021] A computer-readable storage medium for storing program code for performing the above-described adaptive clinical diagnostic method for diseases.

[0022] According to specific embodiments provided by the present invention, the present invention discloses the following technical effects: This application provides an adaptive clinical diagnostic method, apparatus, device, and medium for diseases, achieving multidimensional knowledge distillation and adaptive diagnosis. The method not only aligns with the final prediction distribution of the teacher model but also jointly distills its inference chain and intermediate representations, enabling the student model to possess both diagnostic and reasoning capabilities. Based on this, an adaptive weighting mechanism based on teacher uncertainty dynamically adjusts the learning and diagnostic paths according to case complexity, thereby significantly improving diagnostic accuracy and training stability.

[0023] Meanwhile, structured medical logic constraints were implemented: clinical guidelines and medical knowledge graphs were introduced into both the training and reasoning processes of the method, and soft logic consistency constraints were used to standardize the diagnostic logic, ensuring that the results strictly followed the standard clinical pathways and avoiding diagnostic conclusions that contradicted medical common sense or causal relationships, thereby enhancing the reliability and consistency of the diagnostic results.

[0024] Furthermore, it achieves high transparency and clinical interpretability: through reasoning chain distillation and logical rule verification, this method provides traceable diagnostic evidence, ensuring that each conclusion is supported by clear clinical evidence and reasoning. The diagnostic process is transparent and conforms to medical logic, making it easier for medical professionals to understand and trust, and thus easier to promote and apply in actual clinical settings.

[0025] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0026] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly described below. Obviously, the drawings described below are merely some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without any creative effort.

[0027] Figure 1 This is a flowchart of the adaptive clinical diagnosis method for diseases provided in the embodiments of the present invention; Figure 2 This is a flowchart illustrating the process of applying the adaptive clinical diagnosis method for liver disease provided in this embodiment of the invention to the diagnosis of liver disease. Figure 3 This is a schematic diagram of the clinical diagnostic device for diseases provided in an embodiment of the present invention; Figure 4 This is a schematic diagram of a clinical diagnostic device for diseases provided in an embodiment of the present invention. Detailed Implementation

[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention are within the scope of protection of the present invention.

[0029] See Figure 1 This invention provides an adaptive clinical diagnostic method for diseases, such as... Figure 1 As shown, the method may include: S101: Obtain the patient's electronic medical record dataset, divide and standardize the information contained in the electronic medical record dataset, and establish a multimodal feature vector; the electronic medical record dataset includes at least medical history information, imaging examinations, biochemical tests, and physical examination records; S102: Analyze disease diagnosis and treatment guidelines and medical knowledge graphs, transform the diagnosis and treatment logic into a digital reasoning structure, extract key clinical features and causal links, and embed them into student model training in the form of soft logic constraints; in specific implementation, this application embodiment can provide the analysis of the disease diagnosis and treatment guidelines, extract typical rules, and use fuzzy logic to quantify the overall degree of rule satisfaction. During training, if the predictions given by the student model conflict with clinical rules, the following penalty function is used for punishment:

[0030] In the formula: Indicates sample In the rules Below are the prediction results The rule satisfaction function.

[0031] S103: Introducing answer distribution, inference chain, and intermediate representation information provided by the teacher model, the student model learns a unified representation of diagnostic and reasoning abilities through multi-dimensional distillation; and generating an adaptive weight mechanism based on the uncertainty of the teacher model's prediction results. This adaptive weight mechanism is used to dynamically evaluate and stratify case complexity, dividing cases into simple, medium, and complex complexity levels. In specific implementation, this application embodiment can provide that the multi-dimensional distillation includes answer distillation, inference distillation, and representation distillation. Answer distillation includes using KL divergence to make the student model's prediction distribution approximate the teacher model's; inference distillation includes learning the teacher model's results and inference process, including inference chains or attention distributions; representation distillation includes aligning intermediate layer feature representations, enabling the student model to learn the teacher model's internal representations.

[0032] The adaptive weighting mechanism includes a complexity scoring function and an adaptive weight calculation function, wherein the complexity scoring function is expressed by the following formula:

[0033] In the formula: express, express; The adaptive weight calculation function is expressed by the following formula:

[0034] In the formula: This represents the weighting adjustment coefficient, used to control the degree to which uncertainty amplifies the sample weights. Indicates the output distribution of the teacher model entropy, The teacher model represents the category. The predicted probability.

[0035] S104: The student model after distillation is adaptively invoked according to the complexity of different cases to perform reasoning and judgment; for simple cases, the student model is used to directly give a rapid diagnosis; for moderate cases, a structured logic constraint reinforcement model is introduced to constrain the clinical pathway; for complex cases, a multi-level comprehensive analysis is performed by combining the reasoning chain of multi-dimensional distillation and the logic verification of knowledge graph to ensure that the diagnostic conclusion achieves a balance between accuracy, logical consistency and interpretability.

[0036] The final diagnosis is generated through a joint scoring mechanism of probability distribution and rule satisfaction, and a clear explanation of the evidence is provided. The student model outputs a probability distribution for each disease category and a candidate inference chain based on the input electronic medical record features. Based on medical knowledge graphs and clinical guidelines, calculate the degree to which this case meets different diagnostic rules; The student model's predicted probability is combined with the rule satisfaction level to obtain a comprehensive scoring function; After obtaining the comprehensive scores of all candidate diseases, the disease with the highest score is selected as the final diagnosis result; Along with the diagnostic conclusion, a diagnostic evidence report is generated, which includes a positive evidence set and a missing or insufficient evidence set.

[0037] The comprehensive scoring function is expressed by the following formula:

[0038] In the formula: This indicates that the student model is based on the input samples. The following prediction results are The logarithmic probability is used to characterize the model's own response to the diagnostic results. confidence level This represents the rule weighting coefficient, used to control the strength of the influence of clinical rule constraints on the overall score. This represents the sample complexity function, used to measure the complexity of the input samples. The complexity or uncertainty of the condition, The logarithm of the rule consistency term represents the prediction result. In the sample The degree of alignment with clinical rules, among which The rule satisfaction level (preferably [0,1]) is represented by a higher value, indicating a greater compliance with medical rules.

[0039] The adaptive clinical diagnosis method for diseases provided in this application constructs a lightweight student model that balances diagnostic accuracy, medical logical consistency, and interpretability of reasoning by combining multidimensional knowledge distillation of the teacher model, logical constraints of the clinical knowledge graph, and adaptive sample weighting based on uncertainty during a single training session. Specifically, this method first digitizes the clinical guidelines for the disease, transforming the diagnostic logic into a differentiable reasoning structure; secondly, during the training phase, it uses the answer distribution, reasoning chain, and intermediate representations of the teacher model to perform joint distillation on the student model; finally, it adjusts the learning intensity of cases with different complexities through an adaptive weighting mechanism to achieve integrated diagnostic optimization that conforms to the clinical pathway, thereby improving the accuracy, reliability, and deployability of the diagnosis.

[0040] The following section uses the diagnosis of liver disease as an example to provide a detailed introduction to the adaptive clinical diagnostic method for diseases provided in this application.

[0041] The purpose of this application is to provide an intelligent liver disease diagnosis method that integrates multidimensional knowledge distillation from a teacher model, structured constraints from clinical guidelines, and adaptive sample weighting based on uncertainty. This method, through deep modeling of liver disease clinical data, not only reproduces the diagnostic capabilities of large models but also learns their reasoning paths and intermediate representations. Under the constraints of a medical knowledge graph, it ensures the transparency and consistency of diagnostic logic, thereby improving the accuracy and efficiency of diagnosing complex cases. Simultaneously, by adaptively adjusting the learning difficulty of different cases, the method enables the diagnostic model to possess stronger stability and generalization ability while maintaining interpretability, thus enhancing clinicians' trust in the diagnostic results and facilitating deployment in resource-constrained hospital environments.

[0042] This method overcomes the limitations of traditional knowledge distillation, which only aligns the output distribution. It innovatively introduces a multi-dimensional distillation mechanism, incorporating answer distribution, reasoning chains, and intermediate representations, into student model training. Simultaneously, it introduces logical consistency constraints based on clinical knowledge graphs to ensure that diagnostic results conform to medical reasoning logic. Furthermore, it achieves adaptive sample weighting by measuring the uncertainty of teacher predictions, enabling the student model to dynamically adjust its learning intensity according to case complexity. Specifically, this method first transforms standardized diagnostic logic into differentiable training constraints by automatically parsing clinical guidelines and constructing a medical knowledge graph. Next, it leverages the knowledge guidance of a large-scale teacher model in the answer and reasoning dimensions to jointly optimize the multi-dimensional distillation loss of the student model. Finally, it enhances the learning of complex cases through an adaptive weighting mechanism, ensuring that diagnostic results possess accuracy, logical consistency, and interpretability, thereby providing reliable and personalized diagnostic recommendations for clinicians.

[0043] like Figure 2 As shown, this adaptive clinical diagnostic method for liver diseases based on clinical guideline structural constraints and data-driven distillation includes the following steps: Step 1: Electronic Medical Record Data Input and Preprocessing. The method first reads in an electronic medical record dataset containing patient medical history, imaging examinations, biochemical tests, and physical examination records. This data is then segmented and standardized, and multimodal feature vectors are established to lay the foundation for subsequent knowledge distillation and inference analysis.

[0044] Given a patient's electronic medical record, a multimodal input dataset is formed through the cleaning, standardization, and feature processing of raw clinical data, which can be used for knowledge distillation and logical modeling.

[0045] Data parsing and cleaning: Read electronic medical record data containing patient medical history, imaging examinations, laboratory test results and physical examination information, and parse unstructured text into structured features using a medical language model.

[0046] Numerical normalization and unit standardization: To ensure the comparability of inspection results from different sources, continuous indicators (such as AFP, ALT, etc.) are standardized or normalized.

[0047] Missing and outlier handling: Normalization and missing value imputation methods are used to ensure that different features are comparable under a unified scale, and to provide consistent input for subsequent distillation and inference.

[0048] Multimodal embedding representation: Mapping medical history texts, laboratory data, and imaging features into a unified vector representation through an embedding model:

[0049] Step 2: Clinical Knowledge Structure Modeling. The method involves automatically parsing liver disease treatment guidelines and medical knowledge graphs, transforming the diagnostic logic into a digital reasoning structure, extracting key clinical features and causal links, and embedding them into the model for training in the form of soft logic constraints. This stage forms a differentiable clinical logic chain, providing a standardized basis for the diagnostic process and enabling continuous iteration and optimization with the input of clinical data. Ultimately, it constructs a structured reasoning module that integrates standard treatment pathways.

[0050] To ensure that diagnostic results are not only data-driven but also constrained by medical logic, the model must possess clinical interpretability and credibility during the reasoning process.

[0051] Rule Construction: Typical rules were extracted by analyzing liver disease diagnosis and treatment guidelines. For example: significantly elevated AFP + arterial phase enhancement + portal venous phase washout. Primary liver cancer is highly suspected; Significantly elevated levels + positive autoantibodies It tends to be autoimmune hepatitis. Formalized as:

[0052] in, This indicates the patient's level of satisfaction with a particular characteristic. This indicates the credibility of the candidate diagnostic results.

[0053] Satisfaction Calculation: Fuzzy logic is used to quantify the overall satisfaction level of the rules. For example, for the rule regarding primary liver cancer, if all three conditions are simultaneously highly satisfied, then... It will be close to 1, otherwise it will decrease.

[0054]

[0055] in, This represents the aggregation function in fuzzy logic.

[0056] Logical consistency loss: During training, if the student model's predictions conflict with clinical rules, it will be penalized.

[0057] Step 3: Multidimensional Knowledge Distillation and Adaptive Classification. The method incorporates answer distribution, reasoning chains, and intermediate representation information provided by the teacher model to guide the student model in learning a unified representation of diagnostic and reasoning abilities. Simultaneously, an adaptive weighting mechanism is designed based on the uncertainty of the teacher's prediction results to dynamically evaluate and stratify case complexity, automatically classifying cases into simple, medium, and complex levels. This determines the focus of learning at different stages, avoiding over-reliance on a single feature or distribution.

[0058] By employing multi-dimensional distillation, the student model learns diagnostic results, reasoning processes, and intermediate representations. Simultaneously, uncertainty is incorporated to adjust learning weights, ensuring both efficient and stable training. Specifically, the teacher model provided in this application can be the Qwen-14B large language model, and the student model can be the Qwen-4B large language model.

[0059] Answer distillation: Using KL divergence to make the prediction distribution of the student model approximate that of the teacher model:

[0060] Inference distillation: learning not only the learning outcomes, but also the teacher's reasoning process, including the reasoning chain or attention distribution.

[0061] Characteristic distillation: Aligning intermediate layer feature representations to enable students to learn the teacher's internal representations:

[0062] Construction of combined distillation loss:

[0063] Case complexity stratification: The classification agent establishes a complexity scoring function by analyzing combinations of different features.

[0064] in, It is a scoring function derived by learning from a large amount of case data, aiming to comprehensively consider the impact of various features on the complexity of the disease.

[0065] Ultimately, the classification agent categorizes the illness into three levels: simple, moderate, and complex, using the following formula:

[0066] in, and It is a threshold automatically learned from historical case data, used to distinguish between conditions of different complexities.

[0067] Step 4: Collaborative Analysis and Comprehensive Diagnosis. In the final diagnosis stage, the method adaptively calls upon the distilled student model for reasoning and judgment based on the complexity of different cases. For simple cases, the student model directly provides a rapid diagnosis; for moderate cases, structured logical constraints are introduced to strengthen the model's adherence to the clinical pathway; and for complex cases, a multi-level comprehensive analysis is conducted by combining the reasoning chain of multi-dimensional distillation with knowledge graph logical verification to ensure that the diagnostic conclusions achieve a balance between accuracy, logical consistency, and interpretability.

[0068] The final diagnosis is generated by a combined scoring mechanism of probability distribution and rule satisfaction, and clear evidence is provided to make the results both accurate and interpretable, which is convenient for doctors to refer to and apply in clinical practice.

[0069] Preliminary prediction: Based on the input electronic medical record features, the student model outputs a probability distribution for each disease category, and a candidate inference chain:

[0070] Rule satisfaction calculation: The system will calculate the degree of satisfaction of the case with different diagnostic rules based on the medical knowledge graph and clinical guidelines.

[0071] Joint scoring: Combining the student model's predicted probability with the rule satisfaction level, a comprehensive scoring function is obtained:

[0072] This means that the final diagnosis is not solely determined by model probabilities, but rather by the combined effect of model predictions and clinical rules. If the scores for multiple diseases are close, the system will output a Top-k candidate list and prompt the doctor to conduct further examinations for confirmation.

[0073] After obtaining the comprehensive score of all candidate diseases Then, the system selects the disease with the highest score as the final diagnosis:

[0074] Explanation and Output: Along with the diagnostic conclusion, the system automatically generates a "Diagnostic Basis Report," which includes a set of positive evidence. and missing or insufficient evidence sets These explanations will be consistent with the chain of reasoning. By outputting the information together, doctors can see why the system reached the diagnosis and what information is still insufficient. This transparent evidence alignment mechanism helps doctors trust the system's conclusions rather than accepting the results as a "black box."

[0075] In summary, the adaptive clinical diagnosis method for diseases provided in this application achieves multidimensional knowledge distillation and adaptive diagnosis: this method not only aligns with the final prediction distribution of the teacher model, but also jointly distills its inference chain and intermediate representations, enabling the student model to possess both diagnostic and reasoning capabilities. Based on this, an adaptive weighting mechanism based on teacher uncertainty dynamically adjusts the learning and diagnostic paths according to case complexity, thereby significantly improving the accuracy of diagnosis and the stability of training.

[0076] Meanwhile, structured medical logic constraints were implemented: clinical guidelines and medical knowledge graphs were introduced into both the training and reasoning processes of the method, and soft logic consistency constraints were used to standardize the diagnostic logic, ensuring that the results strictly followed the standard clinical pathways and avoiding diagnostic conclusions that contradicted medical common sense or causal relationships, thereby enhancing the reliability and consistency of the diagnostic results.

[0077] Furthermore, it achieves high transparency and clinical interpretability: through reasoning chain distillation and logical rule verification, this method provides traceable diagnostic evidence, ensuring that each conclusion is supported by clear clinical evidence and reasoning. The diagnostic process is transparent and conforms to medical logic, making it easier for medical professionals to understand and trust, and thus easier to promote and apply in actual clinical settings.

[0078] See Figure 3 This application embodiment can also provide a clinical diagnostic device for diseases, such as... Figure 3 As shown, the device for performing the above-described adaptive clinical diagnosis method for diseases may include: The electronic medical record data input and preprocessing unit 301 is used to acquire the patient's electronic medical record dataset, divide and standardize the information contained in the electronic medical record dataset, and establish a multimodal feature vector; the electronic medical record dataset includes at least medical history information, imaging examinations, biochemical tests, and physical examination records; Clinical knowledge structure modeling unit 302 is used to analyze disease diagnosis and treatment guidelines and medical knowledge graphs, transform the diagnosis and treatment logic in them into digital reasoning structures, extract key clinical features and causal links, and embed them into student model training in the form of soft logic constraints; The multidimensional knowledge distillation and adaptive classification unit 303 is used to introduce the answer distribution, reasoning chain and intermediate representation information provided by the teacher model, and guide the student model to learn the unified representation of diagnostic ability and reasoning ability through multidimensional distillation; and generate an adaptive weight mechanism based on the uncertainty of the prediction results of the teacher model. The adaptive weight mechanism is used to dynamically evaluate and stratify the case complexity, and divide the cases into different complexity levels of simple, medium and complex. The collaborative analysis and comprehensive diagnosis unit 304 is used to adaptively call the distilled student model according to the complexity of different cases for reasoning and judgment. For simple cases, the student model is used to directly provide a rapid diagnosis. For moderate cases, a structured logic constraint reinforcement model is introduced to constrain the clinical pathway. For complex cases, a multi-level comprehensive analysis is performed by combining the reasoning chain of multi-dimensional distillation and the logical verification of knowledge graph to ensure that the diagnostic conclusion achieves a balance between accuracy, logical consistency and interpretability.

[0079] This application embodiment can also provide a clinical diagnostic device for a disease, the device including a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is used to execute the steps of the adaptive clinical diagnosis method for diseases described above according to the instructions in the program code.

[0080] like Figure 4 As shown in the embodiment of this application, a clinical diagnostic device for diseases may include: a processor 10, a memory 11, a communication interface 12, and a communication bus 13. The processor 10, the memory 11, and the communication interface 12 all communicate with each other through the communication bus 13.

[0081] In this embodiment, the processor 10 may be a central processing unit (CPU), a graphics processing unit (GPU), an application-specific integrated circuit, a digital signal processor, a field-programmable gate array, or other programmable logic devices.

[0082] The processor 10 can call programs stored in the memory 11. Specifically, the processor 10 can execute operations in the embodiments of the adaptive disease clinical diagnosis method.

[0083] The memory 11 is used to store one or more programs. The programs may include program code, which includes computer operation instructions. In this embodiment, the memory 11 stores at least a program for implementing the following functions: The process involves acquiring the patient's electronic medical record dataset, dividing and standardizing the information contained in the electronic medical record dataset, and establishing a multimodal feature vector. The electronic medical record dataset includes at least medical history information, imaging examinations, biochemical tests, and physical examination records. We analyze disease diagnosis and treatment guidelines and medical knowledge graphs, transform the diagnosis and treatment logic into a digital reasoning structure, extract key clinical features and causal links, and embed them into student model training in the form of soft logic constraints. The system introduces answer distribution, reasoning chain, and intermediate representation information provided by the teacher model, and guides the student model to learn a unified representation of diagnostic and reasoning abilities through multi-dimensional distillation. Based on the uncertainty of the prediction results of the teacher model, an adaptive weighting mechanism is generated. This adaptive weighting mechanism is used to dynamically evaluate and stratify the complexity of cases, dividing cases into different complexity levels: simple, medium, and complex. The student model is adaptively invoked based on the complexity of different cases for reasoning and judgment. For simple cases, the student model directly provides a rapid diagnosis. For moderate cases, a structured logic constraint reinforcement model is introduced to constrain the clinical pathway. For complex cases, a multi-level comprehensive analysis is performed by combining the reasoning chain of multi-dimensional distillation with knowledge graph logic verification to ensure that the diagnostic conclusion achieves a balance between accuracy, logical consistency and interpretability.

[0084] In one possible implementation, the memory 11 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function (such as file creation or data read / write). The data storage area may store data created during use, such as initialization data.

[0085] In addition, memory 11 may include high-speed random access memory, and may also include non-volatile memory, such as at least one disk storage device or other volatile solid-state storage device.

[0086] Communication interface 12 can be an interface for a communication model, used to connect with other devices or systems.

[0087] Of course, it should be noted that, Figure 4 The structure shown does not constitute a limitation on the clinical diagnostic device for diseases in the embodiments of this application. In practical applications, the clinical diagnostic device for diseases may include more than Figure 4 More or fewer components as shown, or combinations of certain components.

[0088] This application embodiment may also provide a computer-readable storage medium for storing program code for performing the steps of the above-described adaptive clinical diagnosis method for diseases.

[0089] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0090] As can be seen from the above description of the embodiments, those skilled in the art can clearly understand that this application can be implemented by means of software plus necessary general-purpose hardware platforms. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in various embodiments or some parts of the embodiments of this application.

[0091] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple, and relevant parts can be referred to the descriptions in the method embodiments. The systems and system embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0092] The above description is merely a preferred embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention are included within the scope of protection of the present invention.

Claims

1. An adaptive clinical diagnostic method for diseases, characterized in that, include: The process involves acquiring the patient's electronic medical record dataset, dividing and standardizing the information contained in the electronic medical record dataset, and establishing a multimodal feature vector. The electronic medical record dataset includes at least medical history information, imaging examinations, biochemical tests, and physical examination records. We analyze disease diagnosis and treatment guidelines and medical knowledge graphs, transform the diagnosis and treatment logic into a digital reasoning structure, extract key clinical features and causal links, and embed them into student model training in the form of soft logic constraints. The system introduces answer distribution, reasoning chain, and intermediate representation information provided by the teacher model, and guides the student model to learn a unified representation of diagnostic and reasoning abilities through multi-dimensional distillation. Based on the uncertainty of the prediction results of the teacher model, an adaptive weighting mechanism is generated. This adaptive weighting mechanism is used to dynamically evaluate and stratify the complexity of cases, dividing cases into different complexity levels: simple, medium, and complex. The student model is adaptively invoked based on the complexity of different cases for reasoning and judgment. For simple cases, the student model directly provides a rapid diagnosis. For moderate cases, a structured logic constraint reinforcement model is introduced to constrain the clinical pathway. For complex cases, a multi-level comprehensive analysis is performed by combining the reasoning chain of multi-dimensional distillation with knowledge graph logic verification to ensure that the diagnostic conclusion achieves a balance between accuracy, logical consistency and interpretability.

2. The adaptive clinical diagnostic method for diseases according to claim 1, characterized in that, The disease diagnosis and treatment guidelines were analyzed to extract typical rules; fuzzy logic was used to quantify the overall degree of rule satisfaction. During training, if the predictions given by the student model conflict with clinical rules, the following penalty function is used for punishment: In the formula: Indicates sample In the rules Below are the prediction results The rule satisfaction function.

3. The adaptive clinical diagnostic method for diseases according to claim 1, characterized in that, The multidimensional distillation includes answer distillation, reasoning distillation, and representation distillation.

4. The adaptive clinical diagnostic method for diseases according to claim 3, characterized in that, The answer distillation includes using KL divergence to make the prediction distribution of the student model approximate that of the teacher model; the inference distillation includes learning the results of the teacher model and learning the inference process of the teacher model, including inference chains or attention distributions; the representation distillation includes aligning intermediate layer feature representations so that the student model learns the internal representations of the teacher model.

5. The adaptive clinical diagnostic method for diseases according to claim 1, characterized in that, The adaptive weighting mechanism includes a complexity scoring function and an adaptive weight calculation function, wherein the complexity scoring function is expressed by the following formula: In the formula: express, express; The adaptive weight calculation function is expressed by the following formula: In the formula: This represents the weighting adjustment coefficient. Indicates the output distribution of the teacher model entropy, The teacher model represents the category. The predicted probability.

6. The adaptive clinical diagnostic method for diseases according to claim 1, characterized in that, The final diagnosis is generated through a joint scoring mechanism of probability distribution and rule satisfaction, and a clear explanation of the evidence is provided. The student model outputs a probability distribution for each disease category and a candidate inference chain based on the input electronic medical record features. Based on medical knowledge graphs and clinical guidelines, calculate the degree to which this case meets different diagnostic rules; The student model's predicted probability is combined with the rule satisfaction level to obtain a comprehensive scoring function; After obtaining the comprehensive scores of all candidate diseases, the disease with the highest score is selected as the final diagnosis result; Along with the diagnostic conclusion, a diagnostic evidence report is generated, which includes a positive evidence set and a missing or insufficient evidence set.

7. The adaptive clinical diagnostic method for diseases according to claim 6, characterized in that, The comprehensive scoring function is expressed by the following formula: In the formula: This indicates that the student model is based on the input samples. The following prediction results are The logarithmic probability, Indicates the rule weight coefficient. Represents the sample complexity function. The logarithm of the rule consistency term represents the prediction result. In the sample The degree of alignment with clinical rules, This indicates the degree of rule satisfaction.

8. A clinical diagnostic device for a disease, characterized in that, The apparatus for performing the adaptive clinical diagnosis method for diseases according to any one of claims 1-7, the apparatus comprising: The electronic medical record data input and preprocessing unit is used to acquire the patient's electronic medical record dataset, divide and standardize the information contained in the electronic medical record dataset, and establish a multimodal feature vector; the electronic medical record dataset includes at least medical history information, imaging examinations, biochemical tests, and physical examination records; The clinical knowledge structure modeling unit is used to analyze disease diagnosis and treatment guidelines and medical knowledge graphs, transform the diagnosis and treatment logic into digital reasoning structures, extract key clinical features and causal links, and embed them into student model training in the form of soft logic constraints. A multidimensional knowledge distillation and adaptive classification unit is used to introduce the answer distribution, reasoning chain and intermediate representation information provided by the teacher model. Through multidimensional distillation, the student model is guided to learn a unified representation of diagnostic ability and reasoning ability. An adaptive weighting mechanism is generated based on the uncertainty of the prediction results of the teacher model. The adaptive weighting mechanism is used to dynamically evaluate and stratify the case complexity, classifying cases into different complexity levels of simple, medium and complex. The collaborative analysis and comprehensive diagnosis unit is used to adaptively call the distilled student model according to the complexity of different cases for reasoning and judgment. For simple cases, the student model is used to directly provide a rapid diagnosis. For moderate cases, a structured logic constraint reinforcement model is introduced to constrain the clinical pathway. For complex cases, a multi-level comprehensive analysis is performed by combining the reasoning chain of multi-dimensional distillation and the logical verification of knowledge graph to ensure that the diagnostic conclusion achieves a balance between accuracy, logical consistency and interpretability.

9. A clinical diagnostic device for a disease, characterized in that, The device includes a processor and a memory: The memory is used to store program code and transmit the program code to the processor; The processor is configured to execute the adaptive clinical diagnostic method for diseases according to any one of claims 1-7 according to the instructions in the program code.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium is used to store program code for performing the adaptive clinical diagnostic method for diseases according to any one of claims 1-7.

Images