Multi-dimension-based reasoning and interaction system

Abstract

The invention discloses a reasoning and interaction system based on multiple dimensions. According to the inference system based on multiple dimensions, the complete process from multi-source data acquisition to inference result output is realized. The acquisition module is responsible for collecting out-hospital examination information and in-hospital examination information of a patient. And the pruning module predicts a targeted set by utilizing the interactively collected patient information and a preset diagnosis knowledge base, dynamically prunes the full-amount question-examination thinking sub-trees according to the targeted set, and optimizes the pruned question-examination thinking sub-trees. And the feature extraction module extracts key features from the interactively collected out-of-hospital and in-hospital question examination information based on the optimized question examination thinking sub-tree. And finally, the inference module inputs the key features into a pre-trained diagnosis model, and the model is combined with a Bayesian algorithm, a deep learning algorithm and a clinical decision consensus to output a final inference result. The problem that dynamic adaptation and efficient and accurate reasoning of multi-dimensional data cannot be achieved on a task set with specific requirements in the prior art is effectively solved.

Description

Technical Field

[0001] This invention relates to the field of reasoning, and more particularly to a multi-dimensional reasoning and interaction system. Background Technology

[0002] With the rapid development of artificial intelligence (AI) technology, its applications are becoming increasingly widespread across various fields, including healthcare. AI applications in healthcare primarily focus on data analysis, image recognition, and natural language processing, aiming to improve the efficiency and quality of medical services. Against this backdrop, multi-dimensional reasoning systems, as an emerging technological approach, are gradually gaining attention. These systems integrate multiple data sources and utilize complex algorithms for reasoning and decision-making to achieve more efficient information processing and more accurate results.

[0003] However, existing multidimensional reasoning systems still have shortcomings in practical applications to specific task sets, especially in terms of dynamic adjustment of the reasoning process, accuracy and reliability of the model, and interpretability and understandability of the results.

[0004] Existing multi-dimensional reasoning systems typically employ fixed reasoning paths and lack the ability to dynamically adjust the reasoning process based on real-time data. This results in potentially lagging or inaccurate inference outcomes when processing complex and dynamically changing data.

[0005] Existing multi-dimensional reasoning systems often rely on a single algorithm or model when processing multi-source data, making it difficult to capture the complex relationships between data from different dimensions, resulting in insufficient accuracy and reliability of the reasoning results.

[0006] Existing multi-dimensional reasoning systems often lack detailed explanations of the reasoning process and basis when outputting results, making it difficult for users to understand and trust the results provided by the system.

[0007] These shortcomings prevent existing technologies from dynamically adapting to multi-dimensional data and reasoning accurately and efficiently. Summary of the Invention

[0008] This invention provides a multi-dimensional reasoning and interaction system to solve the problem in the prior art that it is impossible to dynamically adapt to multi-dimensional data and perform efficient and accurate reasoning.

[0009] Firstly, this application provides a multi-dimensional reasoning system, including:

[0010] The acquisition module is used to acquire patient outpatient and inpatient medical records.

[0011] The pruning module is used to predict a target set based on patient information and a preset diagnostic knowledge base; according to the target set, the preset full-scale question-and-answer thinking tree is dynamically pruned to obtain a pruned question-and-answer thinking subtree, and the question-and-answer thinking subtree is optimized.

[0012] The feature extraction module is used to extract the first feature from the outpatient and inpatient inquiry information based on the optimized inquiry thinking subtree.

[0013] The reasoning module is used to input the first feature into a preset reasoning diagnostic model so that the reasoning diagnostic model outputs a reasoning result; wherein, the reasoning diagnostic model integrates Bayesian algorithm, deep learning algorithm and clinical decision consensus.

[0014] The acquisition module of this application comprehensively collects patients' outpatient and inpatient medical records, ensuring data integrity and diversity. Next, the pruning module uses a pre-set diagnostic knowledge base and patient information to predict the target set and dynamically prunes the entire medical record subtree accordingly, effectively reducing unnecessary medical records and improving the targeting and efficiency of reasoning. The feature extraction module extracts the first feature from multi-source data based on the optimized medical record subtree, further optimizing the data processing flow. Finally, the reasoning module inputs the first feature into the reasoning diagnostic model trained using Bayesian algorithms, deep learning algorithms, and clinical decision consensus, outputting accurate reasoning results. This process not only improves the accuracy and reliability of reasoning but also enhances the system's adaptability and flexibility through dynamic adjustment and multi-dimensional feature fusion, providing users with more efficient and accurate reasoning services. It effectively solves the problem in existing technologies of being unable to dynamically adapt to multi-dimensional data and perform efficient and accurate reasoning.

[0015] Furthermore, the acquisition module acquires the patient's outpatient and inpatient medical records, specifically as follows:

[0016] The acquisition module obtains the patient's medical history description and self-examination results as outpatient information through a preset interactive interface;

[0017] Based on a pre-set medical database, patient test reports and imaging examination reports are obtained as in-hospital consultation information.

[0018] This application's multi-dimensional reasoning system comprehensively collects patients' outpatient and inpatient examination information through an acquisition module, ensuring data integrity and diversity. Specifically, the acquisition module obtains patients' medical history descriptions and self-examination results as outpatient examination information through a pre-set interactive interface. This data reflects the patient's subjective discomfort and basic health status. Simultaneously, the acquisition module also obtains patients' laboratory reports and imaging examination reports from a pre-set medical database as inpatient examination information. This data provides objective clinical indicators obtained from professional examinations. By integrating these two types of information, the system can gain a more comprehensive understanding of the patient's overall health, providing a multi-dimensional data foundation for subsequent reasoning and diagnosis. This process not only improves the accuracy and reliability of reasoning but also enhances the system's adaptability and flexibility by incorporating real-world application scenarios, providing users with more efficient and accurate reasoning services.

[0019] Furthermore, the pruning module predicts a target set based on patient information and a pre-defined diagnostic knowledge base; specifically:

[0020] The pruning module uses a Bayesian algorithm to match the diagnostic knowledge base with patient information, generating a target set containing multiple prediction probabilities;

[0021] The preset diagnostic knowledge base includes disease entries, corresponding symptoms and signs, auxiliary examinations, diagnostic criteria, differential diagnosis logic, and clinical guidelines; the patient information includes age, gender, and interview information.

[0022] This application's multi-dimensional reasoning system significantly improves the efficiency and accuracy of the reasoning process through a pruning module. The pruning module first uses a Bayesian algorithm to match pre-defined diagnostic experience with patient baseline information, generating a target set containing multiple prediction probabilities. The pre-defined diagnostic experience covers historical diagnostic cases and clinical guidelines from experts, while the patient baseline information includes key data such as age, gender, past medical history, and family medical history. Based on this target set, the system dynamically prunes the pre-defined full-scale question-and-answer subtree, removing questions with low relevance to the current diagnosis, thus obtaining a more concise and efficient pruned question-and-answer subtree. This process not only reduces unnecessary question-and-answer steps and minimizes the waste of medical resources but also improves the targeting and efficiency of reasoning, enabling the system to perform subsequent diagnostic reasoning more quickly and accurately, providing patients with higher-quality medical services.

[0023] Furthermore, based on the target set, the preset full query thinking tree is dynamically pruned to obtain a pruned query thinking subtree, and the query thinking subtree is then optimized, specifically as follows:

[0024] Based on a pre-defined diagnostic knowledge base, a conditional probability matrix and a correlation matrix are constructed to generate a Bayesian network graph.

[0025] The Bayesian network graph is updated according to a preset time decay factor;

[0026] Based on the target set, using the updated Bayesian network graph, symptom nodes with disease relevance higher than a preset threshold are retained to obtain an initial pruned tree;

[0027] Based on the initial pruned tree, retain the first path with a posterior probability higher than a preset threshold, calculate the cost function of each first path, and select the path with the best cost-effectiveness as the second path.

[0028] Based on the initial pruned tree and the second path, the pruned question-and-answer subtree is obtained;

[0029] The question-and-answer subtree is optimized, including restructuring and merging, deleting duplicates, and adjusting the structural order.

[0030] This application's multi-dimensional reasoning system significantly improves the efficiency and accuracy of the inquiry-based subtree through a dynamic pruning mechanism. First, the system constructs a conditional probability matrix and a correlation matrix based on pre-defined clinical guidelines, generating a Bayesian network graph, providing a theoretical basis for dynamic pruning. Then, a pre-defined time-series decay factor is introduced to update the Bayesian network graph to adapt to changes in time-series data, ensuring the model's timeliness and accuracy. Next, based on the target set, the updated Bayesian network graph is used to screen symptom nodes with pathological relevance exceeding a pre-defined threshold, constructing an initial pruned tree. Building upon this, the system further optimizes, retaining only the first path with a posterior probability exceeding the pre-defined threshold, and selecting the most cost-effective path as the second path through cost function evaluation. Finally, the initial pruned tree and the second path are combined to generate the pruned inquiry-based subtree. This process not only simplifies inquiry items and reduces unnecessary examinations but also improves the targeting and efficiency of diagnosis, while optimizing the use of medical resources through cost-benefit analysis.

[0031] Furthermore, the first feature is extracted from the outpatient and inpatient inquiry information based on the optimized inquiry thinking subtree, specifically as follows:

[0032] Feature extraction rules are generated based on the optimized question-and-answer thinking subtree;

[0033] Based on the preset multimodal processing algorithm, the textual semantic features, imaging lesion features, and speech acoustic features of the outpatient examination information are extracted respectively;

[0034] The semantic features, lesion features, and acoustic features are used as features for outpatient examination.

[0035] Standardize the test values ​​in the in-hospital examination information and structure the imaging reports in the in-hospital examination information;

[0036] Standardized test values ​​and structured imaging reports were used as features for in-hospital review.

[0037] Based on the feature extraction rules, the multi-dimensional features of outpatient and inpatient inquiries are weighted and fused to obtain the first feature containing the spatiotemporal dimension.

[0038] This application's multi-dimensional reasoning system significantly enhances the accuracy and comprehensiveness of reasoning through a refined feature extraction process. First, the system formulates feature extraction rules based on the pruned question-and-answer subtree to ensure the targeted and effective extraction. Next, using a pre-defined multi-modal processing algorithm, it extracts semantic features from textual information, lesion features from imaging information, and acoustic features from speech information from outpatient questions, integrating these features into multi-modal features to enrich the data's dimensionality and depth. Simultaneously, it standardizes laboratory values ​​from inpatient questions and extracts structured information from imaging reports to form structured features, further optimizing data quality. Finally, according to the feature extraction rules, it weightedly fuses the multi-modal features and structured features to generate a first feature containing spatiotemporal dimensions. This process not only fully utilizes the advantages of multi-source data but also improves data usability through standardization and structuring, enabling the system to more accurately capture key diagnostic information, providing a solid data foundation for subsequent reasoning and diagnosis, thereby improving the accuracy and reliability of the reasoning results.

[0039] Furthermore, the reasoning diagnostic model is trained based on a fusion of Bayesian algorithm, deep learning algorithm, and pre-defined clinical decision consensus, specifically as follows:

[0040] Obtain a historical training dataset, which includes patient age, gender, and clinical treatment records;

[0041] A probability propagation model is constructed based on the Bayesian algorithm, and a deep feature inference network is constructed based on the deep learning algorithm.

[0042] According to the preset bidirectional mapping method, the parameters of the probability propagation model and the deep feature inference network are fused to obtain the fused initial inference network.

[0043] The initial inference network is trained based on a preset clinical decision consensus engine, and the preset parameters are corrected during the training process to obtain an inference diagnostic model.

[0044] This application's multi-dimensional reasoning system significantly improves the accuracy and reliability of reasoning diagnostic models by comprehensively utilizing Bayesian algorithms, deep learning algorithms, and clinical decision rules. Specifically, the system first acquires a historical training dataset containing basic patient information, examination data, diagnostic results, and clinical decision records, providing a rich data foundation for model training. Next, a probability propagation model is constructed using a Bayesian algorithm to capture probabilistic relationships within the data; simultaneously, a deep feature inference network is built based on a deep learning algorithm to extract complex features from the data. Through a pre-defined bidirectional mapping method, the parameters of the probability propagation model and the deep feature inference network are fused to form a fused initial inference network, achieving complementarity between the advantages of different models. Finally, the initial inference network is trained using a pre-defined clinical decision rule engine, with pre-defined parameters dynamically adjusted during training to obtain an optimized reasoning diagnostic model. This process not only fully utilizes the advantages of multiple algorithms but also ensures that the model's decisions conform to medical standards through the constraints of clinical decision rules, thereby improving the accuracy and reliability of reasoning diagnosis and providing strong support for medical decision-making.

[0045] Furthermore, the multi-dimensional reasoning system of this application also includes a verification module for verifying the reasoning results, specifically:

[0046] The reasoning result is compared with a preset verification standard;

[0047] When the difference between the reasoning result and the verification standard exceeds a preset threshold, a manual review mechanism is triggered.

[0048] Based on the results of manual review, the reasoning and diagnostic model is optimized through feedback.

[0049] This application's multi-dimensional inference system significantly enhances the accuracy and reliability of inference results by introducing a verification module. The verification module first compares the inference results with preset verification standards, ensuring initial accuracy. When the comparison shows a difference exceeding a preset threshold, the system triggers a manual review mechanism. This mechanism allows professionals to conduct a detailed review of the inference results, further ensuring accuracy. Based on the manual review, the system optimizes the inference diagnostic model, correcting potential errors and improving performance and adaptability. Through this combined automatic and manual verification and optimization process, the system provides more accurate and reliable inference results, offering users higher-quality services.

[0050] Furthermore, the multi-dimensional reasoning system of this application also includes a result interpretation module for interpreting and visualizing the reasoning results, specifically:

[0051] The reasoning results are analyzed, the reasoning basis is extracted, and a reasoning explanation report is generated;

[0052] The reasoning and explanation report will be visualized on a pre-set platform.

[0053] This application's multi-dimensional reasoning system significantly enhances the understandability and transparency of reasoning results by introducing a result interpretation module. The result interpretation module first performs in-depth analysis of the reasoning results, extracting key reasoning evidence and recommended checks, and generates a detailed reasoning interpretation report based on this information. Subsequently, the module displays the reasoning interpretation report visually on a pre-defined platform, enabling users to intuitively understand the reasoning process and results. This process not only helps users better grasp the logic and basis of the reasoning but also enhances user trust in the system, making the reasoning results more readily accepted and applied, thereby improving the overall system's usability and user experience.

[0054] Secondly, this application provides a multi-dimensional interactive system, which includes a server, the multi-dimensional inference system described in the first aspect, and a user terminal.

[0055] The server is used to store patients' outpatient and inpatient medical records.

[0056] The user terminal is used to display the reasoning process and reasoning results output by the reasoning system.

[0057] This application's multi-dimensional interactive system integrates a server, user terminal, and advanced inference system to achieve effective management of medical data and efficient presentation of inference results. The server stores the patient's outpatient and inpatient medical history taking characteristics, ensuring data integrity and accessibility. The user terminal serves as the interactive interface, allowing users to input necessary medical history information and clearly displaying the inference results output by the inference system. This process significantly improves the efficiency and accuracy of medical diagnosis, enabling medical professionals to quickly understand and interpret the patient's condition, thereby making more accurate diagnostic decisions. Simultaneously, this interactive design enhances the user experience and improves the quality of medical services.

[0058] Furthermore, the user terminal is also used to receive feedback information from the user regarding the reasoning result, including the user's confirmation, questioning, or supplementary information regarding the reasoning result;

[0059] The feedback information is sent to the server.

[0060] This application's multi-dimensional interactive system enhances the interactivity and adaptability of the system by strengthening the functionality of the user terminal. The user terminal can not only display the inference results output by the inference system but also receive user feedback on these results, which may include confirmation, questioning, or supplementary information. This mechanism allows users to directly participate in the evaluation and correction of the inference results, enabling the system to adjust and optimize the inference model in a timely manner based on actual usage. By sending user feedback to the server, the system can be updated and improved in real time, thereby improving the accuracy and reliability of inference, better meeting users' actual needs, and enhancing users' trust and satisfaction with the system. Attached Figure Description

[0061] Figure 1 : A schematic diagram of an embodiment of the multi-dimensional reasoning system provided in this application;

[0062] Figure 2 : A schematic diagram of an embodiment of the dynamic pruning of the thinking subtree for the diagnostic process provided in this application;

[0063] Figure 3 : A schematic diagram of an embodiment of the multi-dimensional interactive system provided in this application. Detailed Implementation

[0064] 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 some embodiments of the present invention, and 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.

[0065] Example 1

[0066] Please refer to Figure 1 In order to solve the problem that existing technologies cannot perform accurate and efficient reasoning on multi-dimensional data, this invention provides a multi-dimensional reasoning system.

[0067] In this embodiment, the multi-dimensional reasoning system includes an acquisition module 10, a pruning module 20, a feature extraction module 30, and a reasoning module 40.

[0068] The acquisition module 10 is used to acquire the patient's outpatient and inpatient interview characteristics.

[0069] In a preferred embodiment of this example, the acquisition module 10 acquires data in the following manner:

[0070] External examination information: The patient can obtain multimodal information such as images taken by the patient or photos uploaded by the patient's mobile phone (such as images of eye lesions), text input (medical history description), and voice input (medical history narration) through an interactive interface. For example, the patient may report "blurred vision in the right eye with a foreign body sensation for 3 days" and upload photos of redness and swelling in the eye.

[0071] In-hospital information retrieval: Obtain patients' imaging examination reports (such as fundus OCT), laboratory reports (such as blood routine), and previous treatment recommendations from medical databases, such as extracting structured information such as "fundus OCT shows macular edema".

[0072] This application's multi-dimensional reasoning system comprehensively collects patients' pre- and post-hospital medical records through an acquisition module, ensuring data integrity and diversity. Specifically, the acquisition module obtains patients' self-descriptions and self-examination results as pre-hospital medical records through a pre-defined interactive interface. This data reflects the patient's health status and symptom descriptions in a non-medical environment. Simultaneously, the acquisition module also obtains patients' laboratory reports and imaging examination reports from a pre-defined medical database as post-hospital medical records. This data provides detailed examination results of patients in a medical environment. By integrating these two types of information, the system can gain a more comprehensive understanding of the patient's health status, providing a more accurate data foundation for subsequent reasoning and diagnosis. This process not only improves the accuracy and reliability of reasoning but also enhances the system's adaptability and flexibility through the integration of multi-dimensional data, providing users with more efficient and accurate reasoning services.

[0073] The pruning module 20 is used to predict a target set based on patient information and a preset diagnostic knowledge base; according to the target set, it dynamically prunes the preset full-scale question-and-answer thinking tree to obtain a pruned question-and-answer thinking subtree, and optimizes the question-and-answer thinking subtree.

[0074] In a preferred embodiment of this practice, the step of predicting the target set based on patient information and a preset diagnostic knowledge base specifically involves:

[0075] Based on patient information (age 60, female, painless blurred vision with tearing in both eyes for 3 years), combined with a diagnostic knowledge base, a Top-5 suspected disease set (i.e., target set) is calculated using a Bayesian classification algorithm, such as {cataract, diabetic retinopathy, diabetic optic neuropathy, chronic dacryocystitis, dry eye}.

[0076] Furthermore, the dynamic pruning of the preset full query thinking tree to obtain the pruned query thinking subtree is specifically as follows:

[0077] Constructing a symptom-disease Bayesian network atlas: Based on clinical guidelines, establish a conditional probability matrix P(S|D) (e.g., P(blurred vision|diabetic retinopathy) = 0.85), (D∈{d1,d2,...,dn} is the disease set, S∈{s1,s2,...,sm} is the symptom set); and a symptom association matrix R(si,sj) (e.g., the mutual information I between blurred vision and fundus edema = 0.72). Introduce a time decay factor α(t) = e^(-0.02t) to assign higher weights to the features examined in the past 30 days. The weights here can be designed according to the actual situation, and this application does not impose any restrictions.

[0078] Initial pruning: Based on the predicted disease set D', retain symptom nodes where P(S|D')>θ1 (θ1=0.3);

[0079] Iterative pruning: In each round, retain the path that satisfies max[P(D|E),P(D|E')]>θ2 (θ2=0.15, E is the evidence obtained).

[0080] Pathway optimization: Calculate the clinical pathway cost function C = Σ(wi*ti) for the retained pathways, where ti represents parameters such as examination time / cost, and select the cost-effective pathway, such as prioritizing fundus fluorescein angiography over whole-body PET-CT.

[0081] Verification mechanism: Set the pruning confidence threshold δ = 0.85. When P(D|E) < δ, trigger the manual review mechanism.

[0082] Retain at least k (k≥3) competing diagnostic paths to prevent accidental pruning;

[0083] Implement a pruning compensation strategy: record the information entropy H(S|D) of the pruned node, and retain the key identification node when H(S|D)>3.5.

[0084] More specifically, such as Figure 2 As shown, Figure 2 The processing flow for dynamic subtree pruning of the diagnostic process described in this application mainly includes:

[0085] Information Input (Top-N Diseases Input in Question-Check Thinking Subtree): The user or system inputs the Top-N (such as several common or pending ophthalmological diseases preset by the system) ophthalmological diseases as the starting reference for subsequent diagnostic reasoning.

[0086] Information integration (questioning thinking subtree and patient information integration): Integrates patient information, such as outpatient and inpatient questioning information, to provide targeted and individualized questioning thinking.

[0087] Question review subtree pruning (initial pruning, optimization pruning, iterative pruning): Based on the integrated question review subtree and patient information, the entire question review subtree is initially pruned, the pruned question review subtree is optimized (duplicates, reordering, reconstruction and merging), and then iterative pruning is continuously performed based on the patient information obtained in each round of interaction.

[0088] Feature extraction: The question-and-answer thinking subtree extracts features from both in-hospital and out-of-hospital question-and-answer information.

[0089] Model input (multi-dimensional feature input model): Input the extracted multi-dimensional features (covering medical history, examination data, etc.) into the ophthalmology disease diagnosis model, and let the model perform calculations based on these features.

[0090] Model inference (output based on model inference): The model uses algorithms to infer disease-related results by combining input features, such as disease probability and related factors.

[0091] Results Feedback (Returning Top-N High-Matching Inferences): The Top-N high-matching ophthalmological diseases derived from the inferences are fed back to the user or system to assist experts and patients in clarifying the diagnostic direction, forming a complete diagnostic logic loop of "input-integration-pruning-feature extraction-inference-feedback", thereby improving the efficiency and accuracy of inference.

[0092] The feature extraction module 30 is used to extract the first feature from the outpatient and inpatient inquiry information based on the optimized inquiry thinking subtree.

[0093] In a preferred embodiment of this practice, the extraction of the first feature from the outpatient and inpatient inquiry information based on the optimized inquiry thinking subtree specifically involves:

[0094] It should be noted that the first feature includes the patient's medical history, physical signs, and imaging findings.

[0095] Based on the pruned subtree nodes, rules such as "prioritizing the extraction of lesion features from ocular images and prioritizing the extraction of semantic keywords from symptom descriptions" can be determined.

[0096] Furthermore, the extraction of the first feature from the outpatient examination information is performed according to a preset multimodal processing algorithm. This preset multimodal processing algorithm may include deep learning algorithms, etc. As a preferred embodiment of this practice, for example:

[0097] The semantic feature vector of "blurred vision" is extracted by the BERT model and mapped to the medical semantic space;

[0098] The U-Net model was used to segment lesions in fundus images and extract features such as macular edema area and degree of vascular abnormality.

[0099] The frequency features in the speech are extracted using the MFCC algorithm to identify the time trend of symptom descriptions (such as "symptoms gradually worsen").

[0100] Furthermore, the laboratory values ​​in the in-hospital examination information are standardized, and the structured information is extracted from the imaging reports in the in-hospital examination information.

[0101] For example, the white blood cell count in a routine blood test report is standardized, and the structured information of "macular exudation" in a fundus CT report is extracted.

[0102] Standardized laboratory test values ​​and structured imaging reports are used as structured features;

[0103] Based on the feature extraction rules, multimodal features and structured features are weighted and fused to obtain the first feature containing the spatiotemporal dimension;

[0104] For example, multimodal features and structured features are weighted and fused according to temporal weights (recent inspection weight 0.7, historical record weight 0.3) to generate a 1024-dimensional first feature vector containing spatiotemporal dimensions.

[0105] The reasoning module 40 is used to input the first feature into a preset reasoning diagnostic model so that the reasoning diagnostic model outputs a reasoning result; wherein the reasoning diagnostic model is trained based on a fusion of Bayesian algorithm, deep learning algorithm and clinical decision consensus.

[0106] In a preferred embodiment of this invention, the reasoning diagnostic model is trained based on a fusion of Bayesian algorithm, deep learning algorithm, and clinical decision consensus, specifically as follows:

[0107] Obtain a historical training dataset, which includes basic patient information, examination features, diagnostic results, and clinical decision records;

[0108] A disease-symptom probability propagation model is constructed using a Bayesian algorithm, and a deep feature inference network is constructed using a deep learning algorithm. After bidirectional mapping and fusion, the parameters are corrected by the NCCN guideline rule engine.

[0109] The Bayesian algorithm is used to construct a disease-symptom probability propagation model, specifically as follows:

[0110] Step 1: Construct a dynamic Bayesian network DBN, with nodes containing: D (disease), S (symptom), and T (time series features);

[0111] Step 2: Implement the evidence propagation algorithm: P(D|E) = η·ΠP(Ei|D)·P(D|E_prev);

[0112] Step 3: Introduce a time window mechanism: Perform sliding window probability fusion on the inspection results within the time period from t-Δt to t.

[0113] The construction of a deep feature inference network using deep learning algorithms specifically involves:

[0114] Step 1: Construct a Deep Evidence Network (DEN). The input layer includes: structured features (examination values), unstructured features (image report text), and temporal features (symptom evolution).

[0115] Step 2: Calculate feature correlation using a multi-head attention mechanism:

[0116] Attention(Q,K,V)=softmax(QK^T / √d)V;

[0117] Step 3: Perform uncertainty calibration: Obtain the prediction confidence interval through Monte Carlo Dropout.

[0118] The parameters are corrected by the NCCN Guidelines Rule Engine after the bidirectional mapping is fused, specifically as follows:

[0119] Step 1: Convert NCCN, UpToDate, and other guidelines into production rules; wherein, NCCN, UpToDate, and other guidelines are commonly used guidelines in the medical field.

[0120] Step 2: Trigger an alarm when P(D|E) > 0.7 but does not meet the diagnostic criteria;

[0121] Step 3: Simultaneously calculate the scores for the three dimensions of diagnostic support, differential exclusion, and treatment urgency.

[0122] Furthermore, this application optimizes the output of the inference diagnostic model by establishing a hybrid evaluation function:

[0123] Establish a mixed scoring function: Score(D)=w1*P(D|E)+w2*DEN(D)+w3*KB(D);

[0124] Implement dynamic weight adjustment: automatically adjust w1, w2, w3 (Σwi=1) based on the completeness of evidence;

[0125] When outputting Top-N inference results, ensure that: ① 95% of the probability space is covered ② at least one discriminant is included ③ key missing evidence is marked.

[0126] As a preferred embodiment of this application, the multi-dimensional reasoning system further includes a verification module for verifying the reasoning result, specifically:

[0127] The reasoning result is compared with a preset verification standard;

[0128] When the difference between the reasoning result and the verification standard exceeds a preset threshold, a manual review mechanism is triggered.

[0129] Based on the results of manual review, the reasoning and diagnostic model is optimized through feedback.

[0130] As a preferred embodiment of this invention, the multi-dimensional reasoning system of this application further includes a result interpretation module for interpreting and visualizing the reasoning results, specifically:

[0131] The reasoning results are analyzed, the reasoning basis is extracted, and a reasoning explanation report is generated;

[0132] The reasoning and explanation report will be visualized on a pre-set platform.

[0133] The acquisition module of this application comprehensively collects patients' outpatient and inpatient medical records, ensuring data integrity and diversity. Next, the pruning module uses pre-set diagnostic experience and basic patient information to predict a target set and dynamically prunes the full medical record's thought process subtree accordingly, effectively reducing unnecessary medical records and improving the targeting and efficiency of inference. The feature extraction module extracts the first feature from multi-source data based on the pruned medical record's thought process subtree, further optimizing the data processing flow. Finally, the inference module inputs the first feature into the inference diagnostic model trained using Bayesian algorithms, deep learning algorithms, and clinical decision rules, outputting accurate inference results. This process not only improves the accuracy and reliability of inference but also enhances the system's adaptability and flexibility through dynamic adjustment and multi-dimensional feature fusion, providing users with more efficient and accurate inference services. It effectively solves the problem in existing technologies of being unable to dynamically adapt to multi-dimensional data and perform efficient and accurate inference.

[0134] Example 2

[0135] Please refer to Figure 3 The present invention provides a multi-dimensional interactive system.

[0136] In this embodiment, the multi-dimensional interactive system includes a server S10, a multi-dimensional inference system S20 according to Embodiment 1, and a user terminal S30.

[0137] Server S10 is used to store patients' outpatient and inpatient medical records.

[0138] The multi-dimensional reasoning system S20 receives outpatient and inpatient inquiry information stored on the server, performs reasoning on the outpatient and inpatient inquiry information, generates reasoning results, and sends the reasoning process and the reasoning results to the user terminal.

[0139] User terminal S30 is used to receive the reasoning process and reasoning results of the multi-dimensional reasoning system, and to visualize the reasoning process and reasoning results.

[0140] Furthermore, the user terminal is also used to receive feedback information from the user on the reasoning result, including the user's confirmation, questioning, or supplementary information on the reasoning result;

[0141] The feedback information is sent to the server.

[0142] This application's multi-dimensional interactive system integrates a server, user terminal, and advanced inference system to achieve effective management of medical data and efficient presentation of inference results. The server stores patients' outpatient and inpatient medical history taking characteristics, ensuring data integrity and accessibility. The user terminal serves as the interactive interface, allowing users to input necessary medical history taking characteristics and clearly displaying the inference results output by the inference system. This process significantly improves the efficiency and accuracy of medical diagnosis, enabling medical professionals to quickly understand and interpret patients' conditions, thereby making more accurate diagnostic decisions. Simultaneously, this interactive design enhances the user experience and improves the quality of medical services.

[0143] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. In particular, it should be noted that any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention for those skilled in the art.

Claims

1. A multi-dimensional reasoning system, characterized in that, include: The acquisition module is used to acquire patient outpatient and inpatient medical records. The pruning module is used to predict a target set based on patient information and a preset diagnostic knowledge base; according to the target set, the preset full-scale question-and-answer thinking tree is dynamically pruned to obtain a pruned question-and-answer thinking subtree, and the question-and-answer thinking subtree is optimized. The feature extraction module is used to extract the first feature from the outpatient and inpatient inquiry information based on the optimized inquiry thinking subtree. The reasoning module is used to input the first feature into a preset reasoning diagnostic model so that the reasoning diagnostic model outputs a reasoning result; wherein the reasoning diagnostic model is trained based on a fusion of Bayesian algorithm, deep learning algorithm and clinical decision consensus.

2. The multi-dimensional reasoning system according to claim 1, characterized in that, The acquisition module acquires the patient's outpatient and inpatient medical records, specifically: The acquisition module obtains the patient's self-description and self-examination results as outpatient information through a preset interactive interface; Based on a pre-set medical database, patients' laboratory reports and imaging examination reports are obtained as in-hospital consultation information.

3. The multi-dimensional reasoning system according to claim 1, characterized in that, The pruning module predicts a target set based on patient information and a pre-set diagnostic knowledge base, specifically: The pruning module uses a Bayesian algorithm to match the diagnostic knowledge base with patient information, generating a target set containing multiple prediction probabilities; The preset diagnostic knowledge base includes disease entries, corresponding symptoms and signs, auxiliary examinations, diagnostic criteria, differential diagnosis logic, and clinical guidelines; the patient information includes age, gender, and interview information.

4. The multi-dimensional reasoning system according to claim 1, characterized in that, The step involves dynamically pruning the preset full query thinking tree based on the target set to obtain a pruned query thinking subtree, and then optimizing the query thinking subtree, specifically as follows: Based on a pre-defined diagnostic knowledge base, a conditional probability matrix and a correlation matrix are constructed to generate a Bayesian network graph. The Bayesian network graph is updated according to a preset time decay factor; Based on the target set, using the updated Bayesian network graph, symptom nodes with disease relevance higher than a preset threshold are retained to obtain an initial pruned tree; Based on the initial pruned tree, retain the first path with a posterior probability higher than a preset threshold, calculate the cost function of each first path, and select the path with the best cost-effectiveness as the second path. Based on the initial pruned tree and the second path, the pruned question-and-answer subtree is obtained; The question-and-answer subtree is optimized, including restructuring and merging, deleting duplicates, and adjusting the structural order.

5. The multi-dimensional reasoning system according to claim 1, characterized in that, The optimized question-and-answer subtree extracts the first feature from the outpatient and inpatient question-and-answer information, specifically: Feature extraction rules are generated based on the optimized question-and-answer thinking subtree; Based on the preset multimodal processing algorithm, the textual semantic features, imaging lesion features, and speech acoustic features of the outpatient examination information are extracted respectively; The semantic features, lesion features, and acoustic features are used as features for outpatient examination. Standardize the test values ​​in the in-hospital examination information and structure the imaging reports in the in-hospital examination information; Standardized test values ​​and structured imaging reports were used as features for in-hospital review. Based on the feature extraction rules, the multi-dimensional features of outpatient and inpatient inquiries are weighted and fused to obtain the first feature containing the spatiotemporal dimension.

6. The multi-dimensional reasoning system according to claim 1, characterized in that, The reasoning diagnostic model is trained based on a fusion of Bayesian algorithms, deep learning algorithms, and a pre-defined clinical decision consensus, specifically as follows: Obtain a historical training dataset, which includes patient age, gender, and clinical treatment records; A probability propagation model is constructed based on the Bayesian algorithm, and a deep feature inference network is constructed based on the deep learning algorithm. According to the preset bidirectional mapping method, the parameters of the probability propagation model and the deep feature inference network are fused to obtain the fused initial inference network. The initial inference network is trained based on a preset clinical decision consensus engine, and the preset parameters are corrected during the training process to obtain an inference diagnostic model.

7. The multi-dimensional reasoning system according to claim 1, characterized in that, It also includes a verification module for verifying the inference results, specifically: The reasoning result is compared with a preset verification standard; When the difference between the reasoning result and the verification standard exceeds a preset threshold, a manual review mechanism is triggered. Based on the results of manual review, the reasoning and diagnostic model is optimized through feedback.

8. The multi-dimensional reasoning system according to claim 1, characterized in that, It also includes a result interpretation module for interpreting and visualizing the inference results, specifically: The reasoning results are analyzed, the reasoning basis is extracted, and a reasoning explanation report is generated; The reasoning and explanation report will be visualized on a pre-set platform.

9. A multi-dimensional interactive system, characterized in that, Includes a server, the multi-dimensional reasoning system as described in any one of claims 1-8, and a user terminal; The server is used to store patients' outpatient and inpatient medical records. The user terminal is used to display the reasoning process and reasoning results output by the reasoning system.

10. The multi-dimensional interactive system according to claim 9, characterized in that, The user terminal is also used to receive feedback information from the user on the reasoning result, including the user's confirmation, questioning or supplementary information on the reasoning result; The feedback information is sent to the server.

Images