Patient information management system based on artificial intelligence

Abstract

The invention provides a patient information management system based on artificial intelligence, and belongs to the technical field of patient information management, and the system comprises a data processing module, an analysis module and a decision module which are in communication connection in sequence. The data processing module is used for connecting various external medical data sources and performing standardization processing; the analysis module comprises a first analysis sub-module, a second analysis sub-module and an association mapping sub-module which are respectively used for extracting structured features of texts and images by adopting natural language processing and image processing technologies, realizing cross-modal semantic association through dynamic weight distribution and generating patient state information; and the decision-making module performs multi-level path reasoning in combination with the structured medical knowledge network in the knowledge base module to generate clinical auxiliary decision-making information. The system is also integrated with a security control module and a man-machine interaction module, so that data security and decision interpretability are ensured. According to the invention, intelligent integration and clinical decision support of multi-source medical data are realized, and the diagnosis and treatment efficiency and accuracy are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of patient information management, and particularly relates to a patient information management system based on artificial intelligence. BACKGROUND

[0002] In the current medical environment, patient information is usually scattered in multiple heterogeneous systems, such as electronic medical records, image archives, and testing platforms, which leads to difficulties in data integration and serious information island phenomenon. Existing systems are mostly limited to processing a single data type and lack cross-modal semantic association capabilities, making it difficult to fully reflect the patient's health status. In addition, traditional clinical decision support systems mostly rely on static rules, and the reasoning process is not transparent, lacking explainability, and having insufficient dynamic fusion and context awareness capabilities for multi-source data. In terms of data security, most systems have the risk of original data leakage when collaborating across institutions, and cannot meet the compliance requirements of medical privacy protection. Therefore, a patient information management system based on artificial intelligence is proposed to solve the above problems. SUMMARY

[0003] Therefore, the purpose of the present application is to provide a patient information management system based on artificial intelligence to at least solve the above problems.

[0004] The technical solutions adopted by the present application are as follows: A patient information management system based on artificial intelligence, comprising: a data processing module, an analysis module and a decision module connected in sequence, The data processing module is configured to interface with multiple external medical data sources and to standardize the received data; The analysis module is configured to receive and process the standardized data; the analysis module includes an association mapping submodule, which establishes semantic association relationships between medical feature information from different data sources through a dynamic weight distribution method, and generates patient state information based on the semantic association relationships; The decision module is configured to receive the patient state information and generate clinical auxiliary decision information.

[0005] Further, the analysis module further includes a first parsing submodule and a second parsing submodule; The first parsing submodule uses natural language processing technology to perform sequence labeling on text type standardized data objects to extract medical entities and their semantic relationships that meet medical terminology standards, and outputs first structured feature information in the form of entity-relation-entity triplets; The second analysis submodule adopts image processing technology to pre-process, regionally segment and feature extract the image type standardized data object, identify specific medical signs and quantitatively analyze the same, and output second structured feature information conforming to a medical image reporting standard; The association mapping submodule is in communication connection with the first analysis submodule and the second analysis submodule respectively, and is configured to receive the first structured feature information and the second structured feature information.

[0006] Further, the dynamic weight distribution is realized by the following manners: The association mapping submodule constructs a shared feature space, and converts medical entities in the first structured feature information and medical signs in the second structured feature information into feature vectors respectively; Based on context information of a current clinical scene, the correlation weight between feature vectors of different sources is determined through attention calculation; According to the calculated correlation weight, the feature vectors are weighted fused and spatially aligned to generate the patient state information.

[0007] Further, the knowledge base module is further included; The knowledge base module stores a structured medical knowledge network with medical concepts as nodes and causal, diagnostic and treatment logical relationships between the concepts as connecting edges; The decision module is in communication connection with the knowledge base module, and is configured to: extract key medical features from the patient state information; perform semantic matching between the key medical features and medical concept nodes in the knowledge base module; based on the matched nodes, perform multi-level path traversal and reasoning in the structured medical knowledge network; generate reasoning basis according to the traversal and reasoning result, integrate the reasoning basis with a preliminary decision suggestion generated based on the patient state information, and form the clinical auxiliary decision information.

[0008] Further, the decision module performs the following operations when performing multi-level path traversal and reasoning: find all possible paths connecting nodes corresponding to the key medical features and a pre-defined clinical conclusion node in the structured medical knowledge network; based on confidence degrees of relationships between nodes in the paths and path lengths, calculate comprehensive weights of the paths; select a path with the highest comprehensive weight, and take all nodes and connecting edge sequences in the path as the most possible reasoning basis.

[0009] Further, it further comprises a security control module; The security control module is in communication connection with the data processing module, and is used for: When the system needs to cooperate with an external system for analysis, intercepting a request for sending original data outward; Starting a security calculation process, and receiving calculation parameters sent by the external system; In the local system, using the standardized data object to complete a calculation task; Only the calculation result is sent back to the external system after being encrypted.

[0010] Further, it further comprises a man-machine interaction module; The man-machine interaction module is in communication connection with the decision module, and is used for: Receiving the clinical auxiliary decision information and its complete inference basis chain; Generating an interactive visualization interface, wherein the interface at least comprises a decision subject area for displaying decision suggestions, an evidence chain area for dynamically displaying nodes and paths in the inference basis in the form of a graph, and an original evidence area for associatively displaying original data segments from the first analysis submodule and the second analysis submodule; The nodes and paths in the evidence chain area have an interactive linkage function with corresponding data segments in the original evidence area.

[0011] Further, the man-machine interaction module further performs the following operations: Receiving feedback operations of the user on the clinical auxiliary decision information through the interactive visualization interface; Analyzing the feedback operations into confirmation, correction or negation signals of the current decision suggestions; Returning the signals to the associated mapping submodule of the analysis module and the decision module; The associated mapping submodule adjusts the correlation weight calculation method in the dynamic weight distribution according to the correction or negation signal; and the decision module optimizes the medical rule library and algorithm library used by the decision module according to the signal.

[0012] Compared with the prior art, the present application has the following beneficial effects: 1. The standardized access and deep fusion of multi-source heterogeneous medical data are realized, the structured features of text and images are extracted through natural language processing and image processing technology respectively, the data island is broken, and a data basis is provided for comprehensive patient state analysis.

[0013] 2. Through a dynamic weight distribution mechanism, the correlation weights between different features are dynamically adjusted based on clinical context, cross-modal semantic correlation is realized, and patient state information more suitable for clinical practice is generated.

[0014] 3. By combining structured medical knowledge networks with multi-level path reasoning, interpretable clinical decision support information is generated, improving the transparency and credibility of decision-making. Attached Figure Description

[0015] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only preferred embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0016] Figure 1 This is a schematic diagram of the overall structure of a patient information management system based on artificial intelligence proposed in an embodiment of the present invention. Detailed Implementation

[0017] The principles and features of the present invention are described below with reference to the accompanying drawings. The listed embodiments are only used to explain the present invention and are not intended to limit the scope of the present invention.

[0018] Reference Figure 1 This invention provides an artificial intelligence-based patient information management system, comprising: a data processing module, an analysis module, and a decision-making module that are sequentially and communicatively connected. The data processing module is used to connect to various external medical data sources and perform standardized processing on the received data; The analysis module is used to receive and process standardized data; the analysis module includes an association mapping submodule, which establishes semantic associations between medical feature information from different data sources through dynamic weight allocation, and generates patient status information based on the semantic associations. The decision-making module is used to receive the patient status information and generate clinical auxiliary decision-making information.

[0019] Exemplarily, the data processing module can undertake the responsibilities of docking and standardizing multi-source heterogeneous medical data. The data processing module needs to establish a communication connection with external medical data sources such as hospital electronic medical record systems, medical image archiving systems, laboratory testing platforms, and wearable device data interfaces, receive diversified raw medical information including structured data (such as test index values), semi-structured data (such as outpatient medical record texts), and unstructured data (such as medical image raw files), and after data access, the module converts the raw data of different data sources into a unified data model through standardization operations such as data cleaning, format conversion, encoding mapping, and structured reorganization, for example, converts medical image DICOM files of different manufacturers into a standard format, or maps free text descriptions in clinical texts into standardized medical terminologies, thereby eliminating data heterogeneity and providing machine-recognizable standardized data input for subsequent modules.

[0020] The analysis module can receive the standardized data and perform multi-dimensional analysis and dynamic correlation analysis. The analysis module includes a correlation mapping sub-module that realizes semantic correlation modeling of medical features across data sources through a dynamic weight distribution mechanism. The correlation mapping sub-module first extracts features from the standardized data and identifies medical features from different data sources (such as symptom descriptions in text data and lesion morphology in image data). Then, the correlation mapping sub-module dynamically adjusts the correlation weights between different features according to the context information of the current clinical scenario (such as patient complaints and current diagnosis and treatment stage), for example, increases the weight of vital sign indicators in acute illness scenarios and strengthens the weight of long-term follow-up data in chronic disease management scenarios. Through this dynamic weight distribution, the correlation mapping sub-module can establish semantic correlation relationships between medical features from different sources, such as correlating lung ground glass nodule features in images with cough symptom descriptions in electronic medical records, and finally generate patient status information that comprehensively reflects the current health status of the patient. The patient status information can include comprehensive status descriptions after multi-dimensional feature correlation.

[0021] The decision module can receive the patient status information generated by the analysis module and generate clinical auxiliary decision information based on the medical knowledge system. The decision module integrates the multi-dimensional feature correlation results in the patient status information, applies pre-set clinical decision rules or machine learning models, and generates clinical auxiliary decision information including diagnosis suggestions, treatment plan recommendations, prognosis evaluations, etc. The clinical auxiliary decision information is based on the comprehensive analysis results of the patient status information to provide scientific and precise decision support for clinical medical staff and assist in improving diagnosis and treatment efficiency and quality.

[0022] The system realizes the intelligent management of the whole process from the input of original data to the output of clinical decision-making, from the access and standardization processing of multi-source data to the dynamic correlation analysis of cross-data source features, and finally to the intelligent decision-making information available in clinical.

[0023] The analysis module further comprises a first analysis submodule and a second analysis submodule. The first analysis submodule adopts natural language processing technology to perform sequence labeling on text type standardized data objects to extract medical entities and semantic relationships therebetween conforming to medical terminology standards and output first structured feature information in the form of entity-relation-entity triplets. The second analysis submodule adopts image processing technology to perform preprocessing, region segmentation and feature extraction on image type standardized data objects, identify specific medical signs and perform quantitative analysis thereon, and output second structured feature information conforming to medical image report standards. The correlation mapping submodule is in communication connection with the first analysis submodule and the second analysis submodule respectively, and is used to receive the first structured feature information and the second structured feature information.

[0024] For example, the first analysis submodule can perform in-depth analysis on text type standardized data. The first analysis submodule adopts natural language processing technology to perform sequence labeling on input medical text (such as electronic medical records, clinical records, etc.) so as to accurately identify and extract medical entities (such as disease names, symptom manifestations, test indicators, etc.) conforming to medical terminology standards, and at the same time, clarify the semantic correlation relationships (such as cause-effect relationship, accompanying relationship, time sequence relationship, etc.) between entities. Finally, the first structured feature information is output in the form of entity-relation-entity triplets. For example, for the text of “patient complains of intermittent chest pain for two weeks”, the first analysis submodule can extract “chest pain” as the core entity, “complaint” as the relationship attribute, and “patient” as the associated entity to form a standardized triplet structure, ensuring that the text information is converted into structured knowledge understandable by machines.

[0025] The second analysis submodule can process the image type standardized data. The second analysis submodule relies on image processing technology to complete a multi-step analysis process. First, the input medical images (such as CT, MRI, ultrasound images, etc.) are preprocessed, including image enhancement, denoising, standardization, etc., to optimize the image quality. Then, through regional segmentation technology, the target area (such as tumor lesions, organ contours, etc.) is accurately located, and key features (such as morphological features, density values, texture features, etc.) are extracted. Finally, based on the medical image reporting standard, specific medical signs (such as the boundary clarity of nodules, the abnormality of blood vessel running, etc.) are identified and quantitatively analyzed to form the second structured feature information that meets the clinical diagnosis requirements. For example, in a lung CT image, the second analysis submodule can identify the morphological features of lung nodules, quantify their volume, density, etc., and correlate them to the corresponding image reporting standard to generate standardized image feature descriptions.

[0026] The association mapping submodule can receive the first structured feature information (text analysis result) and the second structured feature information (image analysis result) from the two analysis submodules, providing basic data support for subsequent establishment of semantic association relationships across data sources. By integrating the structured features of text and images, the association mapping submodule can lay a multi-dimensional, cross-modal data fusion foundation for generating patient status information, ensuring that medical information from different sources is effectively associated and aligned at the semantic level, thereby supporting more comprehensive patient status analysis and clinical decision-making.

[0027] The dynamic weight distribution is achieved by the following methods: The association mapping submodule constructs a shared feature space, converting the medical entities in the first structured feature information and the medical signs in the second structured feature information into feature vectors, respectively; Based on the context information of the current clinical scenario, the correlation weights between feature vectors from different sources are determined through attention calculation; According to the calculated correlation weights, the feature vectors are weighted and fused and spatially aligned to generate the patient status information.

[0028] For example, constructing a shared feature space can achieve unified expression of cross-modal features. The association mapping submodule can convert structured features from different data sources, such as medical entity triples (entity-relation-entity) output by the first analysis submodule and medical sign quantification features extracted by the second analysis submodule, into feature vectors. Through feature engineering methods, the medical features of different modalities can be made comparable at the semantic level, such as mapping the "chest pain" entity and the "lung nodule sign" to the same measurable position in the vector space, so that they form a computable geometric relationship in space.

[0029] The attention calculation can realize the context-aware weight distribution. Based on the context information of the current clinical scene (such as patient complaints, current diagnosis stage, treatment goals, etc.), the system dynamically calculates the correlation weight between different source feature vectors, so as to simulate the thinking process of the clinician when making decisions: in a specific scene, some features (such as acute chest pain) are given higher weights, and the weights of other features (such as chronic disease history) are correspondingly reduced. The weight distribution is not a fixed value, but is adjusted in real time with the change of the clinical scene, so as to ensure that the feature correlation is always highly matched with the current diagnosis and treatment demand.

[0030] The patient state information is generated by weighted fusion and spatial alignment. According to the calculated correlation weight, the system performs weighted fusion on the feature vectors, so as to ensure the consistency of features from different sources in the semantic space. For example, the "lung nodule" entity in the text feature is spatially aligned with the "nodule edge burr sign" in the image feature, so that they form a close correlation in the shared feature space, and finally the patient state information fused with multi-dimensional medical features is generated, so as to provide more comprehensive state description for subsequent clinical decision making. Through the context-aware weight calculation, the dynamic modeling of feature correlation is realized, so that the generation process of patient state information is closer to the multi-source information integration demand in the actual clinical diagnosis and treatment scene, thereby improving the processing capacity of the system for complex clinical problems.

[0031] The embodiment also includes a knowledge base module; The knowledge base module stores a structured medical knowledge network composed of medical concepts as nodes and causal, diagnostic and treatment logical relationships between concepts as connecting edges; The decision module is also in communication connection with the knowledge base module, and is used for: extracting key medical features from the patient state information; performing semantic matching of the key medical features with the medical concept nodes in the knowledge base module; based on the successfully matched nodes, performing multi-level path traversal and reasoning in the structured medical knowledge network; generating inference basis according to the traversal and reasoning result, and integrating the inference basis with the preliminary decision suggestion generated based on the patient state information to form the clinical auxiliary decision information.

[0032] Exemplarily, the knowledge base module takes a structured medical knowledge network as the core, which is composed of medical concept nodes and logical relationship edges between nodes such as causality, diagnosis, treatment, etc., forming a reasoning medical knowledge system. During the running process of the decision module, first, key medical features are extracted from the patient state information, such as typical symptoms or examination result indicators of a specific disease. Then, these features are semantically matched with the medical concept nodes in the knowledge base to ensure accurate correspondence between the features and the knowledge nodes at the semantic level. After matching, the decision module starts a multi-level path traversal and reasoning process in the structured medical knowledge network, taking the node corresponding to the key medical feature as the starting point and the pre-defined clinical conclusion node (such as diagnosis conclusion, treatment plan suggestion) as the end point, in order to explore all possible logical paths between them. The evaluation of each path needs to consider the confidence of the relationship between nodes (reflecting the reliability of medical logical relationship) and the path length (affecting the reasoning efficiency and result stability), and the optimal path is selected through comprehensive weight calculation. The node sequence and connection edge relationship contained in the optimal path constitute the most possible inference basis. Finally, the decision module integrates the inference basis and the preliminary decision suggestion generated based on the patient state information. The integration process not only includes the verification and supplement of the preliminary suggestion, but also ensures that the decision information contains a complete logical reasoning chain, so that the clinical auxiliary decision not only has the objectivity of data-driven, but also conforms to the logical consistency of the medical knowledge system, thereby providing interpretable and traceable intelligent decision support for medical staff.

[0033] When the decision module performs multi-level path traversal and reasoning, the following operations are performed: In the structured medical knowledge network, find all possible paths connecting the node corresponding to the key medical feature and the pre-defined clinical conclusion node; Based on the confidence of the relationship between nodes in the path and the path length, calculate the comprehensive weight of each path; Select the path with the highest comprehensive weight, and take all the nodes and connection edge sequences passed through by the path as the most possible inference basis.

[0034] For example, in a structured medical knowledge network composed of medical concept nodes (such as disease symptoms, test indicators, treatment plans, etc.) and logical connection edges (such as causal relationships, diagnostic associations, treatment paths, etc.), the decision module first needs to locate the key medical feature nodes corresponding to the patient state information (such as "persistent fever", "white blood cell elevation", etc. specific clinical features), and determine the pre-defined clinical conclusion nodes (such as "bacterial infection" diagnosis conclusion or "antibiotic use" treatment recommendation), then the system will comprehensively search all potential reasoning paths connecting the aforementioned key feature nodes and target conclusion nodes in the network, these paths are composed of consecutive node sequences and connection edges, each path represents a possible logical reasoning chain, for example, the step-by-step derivation process from "fever-white blood cell elevation" to "infectious disease" to "bacterial infection"; In the path evaluation stage, the system needs to consider two core dimensions: one is the confidence between nodes in the path, which reflects the evidence strength or consensus of a specific medical logical relationship (such as the association between "fever and infection") in clinical practice, usually based on authoritative medical guidelines, evidence-based medical research or historical diagnosis and treatment data accumulation; the second is the topological length of the path, that is, the number of nodes passed from the starting feature node to the conclusion node, the shorter the path, the more direct and efficient the reasoning process, the system calculates the comprehensive weight value for each path by integrating these two types of dimension information (such as using a weighted combination strategy of confidence and path length), which is essentially a quantitative expression of the reliability and simplicity of the path. Ultimately, the decision module will select the path with the highest comprehensive weight as the optimal reasoning basis, this selected path not only contains the key node sequence of the conclusion (such as "fever, infectious disease, bacterial infection"), but also contains the logical connection edges between the nodes (such as "fever is a common manifestation of infectious disease", "bacterial infection accounts for a high proportion in infectious diseases", etc.), forming a complete reasoning evidence chain. In this way, the system can dynamically generate clinical auxiliary decision information that not only conforms to medical logic but also has high credibility, ensuring that the reasoning process of the decision recommendation is traceable, verifiable, and adaptable to the dynamic balance needs of evidence strength and reasoning efficiency in different clinical scenarios.

[0035] The embodiment also includes a security control module; The security control module is in communication connection with the data processing module, and is used for: When the system needs to cooperate with an external system for analysis, intercepting a request for sending original data outward; Starting a security computing process and receiving computing parameters sent by the external system; Using the standardized data object to complete the computing task in the local system; Only the computing result is sent back to the external system after being encrypted.

[0036] For example, when the system needs to cooperate with an external medical system for analysis, the security control module first performs data transmission interception operation, that is, it prevents direct transmission of raw medical data without processing, so as to block the risk of sensitive information outflow from the source and ensure that patient privacy data will not be exposed in plaintext form outside the system boundary; then, the security control module starts a security calculation process, which is triggered based on the calculation parameters (such as parameter configuration of a specific analysis algorithm or joint calculation task instruction) sent by the external system. In this process, the system only calls standardized medical data objects to perform calculation tasks in the local environment, for example, through local calculation engine to jointly analyze standardized electronic medical record text and medical image features. This "data does not move, calculation moves" mode ensures that the original data always remains in the local controlled environment, and only the calculation results are allowed to be transmitted outward; in the result transmission stage, the security control module implements encryption processing on the calculation results, encapsulates the result data using an encryption algorithm (such as AES-256 or national encryption algorithm) conforming to the medical information security standard, and sends the encrypted data packet to the external system. This end-to-end encryption transmission mechanism, combined with the previous data interception and local calculation strategy, builds a full-process security protection system from data input to result output, which not only meets the needs of cross-system cooperation for diagnosis and treatment analysis, but also strictly follows the compliance requirements of medical data privacy protection.

[0037] The embodiment also includes a human-computer interaction module; The human-computer interaction module is in communication connection with the decision module, and is configured to: receive the clinical auxiliary decision information and its complete inference basis chain; generate an interactive visualization interface, which at least includes a decision subject area for displaying decision suggestions, an evidence chain area for dynamically displaying nodes and paths in the inference basis in the form of a graph, and an original evidence area for associating and displaying original data segments from the first analysis submodule and the second analysis submodule; The nodes and paths in the evidence chain area and the corresponding data segments in the original evidence area have an interactive linkage function.

[0038] For example, the human-computer interaction module can receive clinical auxiliary decision-making information and its complete inference chain from the decision-making module. This chain contains the entire logical path from patient status information to the final decision recommendation. For example, it follows a reasoning trajectory from key features such as "persistent cough" and "abnormal lung imaging," through intermediate nodes such as "infectious disease" and "pneumonia," ultimately pointing to the recommendation of "antibiotic treatment." Based on the received information, the human-computer interaction module generates an interactive visual interface, which consists of three functional areas working together: the decision-making body area is used to clearly display the core decision recommendation, such as "recommend using cephalosporins." "Antibiotic-like"; the evidence chain area presents the nodes and paths in the inference basis in the form of a dynamic map. The nodes represent medical concepts (such as symptoms, test indicators, and diagnostic conclusions), and the paths represent the logical connections between the nodes (such as causal, diagnostic, and treatment relationships). The map can be dynamically expanded or collapsed according to the user's operation, intuitively reflecting the hierarchical structure of the reasoning chain; the original evidence area displays the original data fragments that support the aforementioned reasoning, such as the text description of "cough lasting for 3 days" in the electronic medical record extracted by the first analysis submodule, or the image feature of "patchy shadow in the lower lobe of the left lung" in the chest CT image identified by the second analysis submodule.

[0039] The interactive linkage function between the evidence chain area and the original evidence area: When a user clicks on a node (such as "abnormal lung imaging") or path (such as the association between "cough → pneumonia") in the evidence chain area, the system automatically highlights the corresponding original data fragment in the original evidence area (such as the annotation of specific abnormal areas in CT images, or the description of cough symptoms in medical records). Conversely, if a user views the original evidence fragment and has questions, they can also trace back to the corresponding node in the evidence chain through reverse operation, thereby achieving bidirectional tracing from abstract reasoning to concrete evidence. The human-computer interaction module not only improves the transparency and explainability of the decision-making process, but also supports users in actively verifying the reliability of reasoning logic, enhancing clinicians' trust and acceptance of auxiliary decision-making information.

[0040] The human-computer interaction module also performs the following operations: Receive feedback from users on the clinical decision support information provided through the interactive visual interface; The feedback operation is interpreted as a confirmation, correction, or rejection signal for the current decision recommendation; The signal is transmitted back to the correlation mapping submodule of the analysis module and the decision module; The correlation mapping submodule adjusts the correlation weight calculation method in the dynamic weight allocation according to the correction or negation signal; the decision module optimizes the medical rule base and algorithm base it uses according to the signal.

[0041] For example, when the user deeply interacts with the system through the interactive visualization interface, multi-dimensional feedback operations can be initiated for clinical auxiliary decision-making information. For example, in the evidence chain area, the user can adjust the reasoning path order by dragging the nodes, circle a specific path to express recognition or doubt, or click on the associated medical image segment or text record segment in the original evidence area to mark data quality; in the decision subject area, the user can directly click the “confirm” button to express complete acceptance of the current decision suggestion, or select the “modify” option to make detailed adjustments to the specific conclusion (such as modifying the diagnosis tendency or supplementing treatment suggestions), or trigger the “deny” operation to question the decision logic. These interactive behaviors are all analyzed by the system as structured feedback signals, i.e., the confirmation signal represents that the current decision logic completely matches the user's experience, the modification signal reflects the user's specific adjustment appeal to the node association strength and path selection logic in the reasoning path, and the denial signal indicates the user's fundamental doubt about the current decision basis.

[0042] After the feedback signal is analyzed, the system will start a two-way optimization process: on the one hand, the correlation mapping submodule will adjust the dynamic weight distribution strategy after receiving the modification or denial signal, for example, when the user repeatedly modifies the correlation weight of a specific medical entity (such as “cough” and “pulmonary infection”), the system will re-evaluate the vector representation of this feature in the shared feature space, dynamically adjust the relevance weight distribution logic of feature vectors from different data sources (such as electronic medical record text and chest CT images) through attention mechanism, and generate more semantic correlation patterns that fit the clinical actual scene for subsequent patient state information; on the other hand, the decision module will optimize its knowledge reasoning unit after receiving the feedback signal, i.e., by updating the logical connection edge confidence parameters in the medical rule library (such as increasing the evidence weight of the “fever-infection” relationship), adjusting the path traversal strategy in the algorithm library (such as preferentially selecting short paths that have been confirmed by the user multiple times), or supplementing new clinical conclusion nodes and reasoning paths, so as to generate more reasoning basis chains that conform to the user's professional judgment in subsequent decisions.

[0043] Construct a dynamic adaptation channel from clinical practice to system optimization: the user's professional experience directly feeds back to the system parameter adjustment through the visualization interface, enabling the system to continuously learn the implicit knowledge in clinical decision-making (such as the preference of a specific department for symptom correlation strength and the special reasoning path of rare cases), gradually improving the individualized adaptation and clinical explainability of decision suggestions, and at the same time, the two-way transmission of feedback signals ensures the collaborative optimization of the analysis module and the decision module, i.e., the weight adjustment of the correlation mapping submodule directly affects the generation quality of the patient state information, while the rule optimization of the decision module feeds back to the reliability of the reasoning path, ultimately forming a virtuous cycle of continuous iteration of decision quality, enabling the system to maintain dynamic adaptation and intelligent evolution ability in complex clinical scenarios.

[0044] The above merely describes preferred embodiments of the present application and is not used to limit the present application, and any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A patient information management system based on artificial intelligence, characterized in that, include: The data processing module, analysis module, and decision-making module are sequentially connected via communication. The data processing module is used to connect to various external medical data sources and perform standardized processing on the received data; The analysis module is used to receive and process standardized data; the analysis module includes an association mapping submodule, which establishes semantic associations between medical feature information from different data sources through dynamic weight allocation, and generates patient status information based on the semantic associations. The decision-making module is used to receive the patient status information and generate clinical auxiliary decision-making information.

2. The system as described in claim 1, characterized in that, The analysis module further includes a first parsing submodule and a second parsing submodule; The first parsing submodule uses natural language processing technology to perform sequence labeling on standardized text data objects to extract medical entities that conform to medical terminology standards and their semantic relationships, and outputs them as first structured feature information. The first structured feature information is represented in the form of entity-relationship-entity triples. The second parsing submodule uses image processing technology to preprocess, segment, and extract features from standardized image data objects, identify specific medical signs and perform quantitative analysis on them, and output second structured feature information that conforms to medical image report standards. The association mapping submodule is communicatively connected to the first parsing submodule and the second parsing submodule, respectively, and is used to receive the first structured feature information and the second structured feature information.

3. The system as described in claim 2, characterized in that, The dynamic weight allocation is achieved in the following way: The association mapping submodule constructs a shared feature space, converting the medical entities in the first structured feature information and the medical signs in the second structured feature information into feature vectors respectively; Based on the contextual information of the current clinical scenario, the correlation weights between feature vectors from different sources are determined through attention calculation. Based on the calculated relevance weights, the feature vectors are weighted, fused, and spatially aligned to generate the patient status information.

4. The system as described in claim 3, characterized in that, It also includes a knowledge base module; The knowledge base module stores a structured medical knowledge network consisting of medical concepts as nodes and causal, diagnostic, and treatment logical relationships between concepts as connecting edges. The decision-making module is also communicatively connected to the knowledge base module, for the following purposes: Extract key medical features from the patient status information; The key medical features are semantically matched with the medical concept nodes in the knowledge base module; Based on the successfully matched nodes, multi-level path traversal and reasoning are performed in the structured medical knowledge network. Based on the traversal reasoning results, inference criteria are generated, and these inference criteria are integrated with preliminary decision suggestions generated based on the patient's status information to form the clinical auxiliary decision information.

5. The system as described in claim 4, characterized in that, When performing multi-level path traversal and reasoning, the decision-making module specifically performs the following operations: In the structured medical knowledge network, find all possible paths connecting the nodes corresponding to the key medical features with the predefined clinical conclusion nodes; The comprehensive weight of each path is calculated based on the confidence level of the relationship between nodes in the path and the path length. The path with the highest overall weight is selected, and the sequence of all nodes and connecting edges traversed by this path is used as the most likely basis for inference.

6. The system as described in claim 1, characterized in that, It also includes a security management module; The security control module is communicatively connected to the data processing module and is used for: When the system needs to collaborate with external systems for analysis, it intercepts requests to send raw data outwards. Initiate the secure computation process and receive computation parameters sent by the external system; Within the local system, the standardized data objects are used to complete the computational task; The calculation results are encrypted before being sent back to the external system.

7. The system as described in claim 1, characterized in that, It also includes a human-computer interaction module; The human-computer interaction module is communicatively connected to the decision-making module and is used for: Receive the clinical decision support information and its complete chain of inferences; Generate an interactive visualization interface, which includes at least a decision subject area for displaying decision suggestions, an evidence chain area for dynamically displaying nodes and paths in the inference basis in the form of a graph, and an original evidence area for displaying original data fragments originating from the first analysis submodule and the second analysis submodule. The nodes and paths in the evidence chain area have interactive linkage functions with the corresponding data fragments in the original evidence area.

8. The system as described in claim 1, characterized in that, The human-computer interaction module also performs the following operations: Receive feedback from users on the clinical decision support information provided through the interactive visual interface; The feedback operation is interpreted as a confirmation, correction, or rejection signal for the current decision suggestion; The signal is transmitted back to the correlation mapping submodule of the analysis module and the decision module; The correlation mapping submodule adjusts the correlation weight calculation method in the dynamic weight allocation according to the correction or negation signal; the decision module optimizes the medical rule base and algorithm base it uses according to the signal.

Images