Intelligent management system for clinical data of medical examination

Abstract

The invention discloses a medical examination clinical data intelligent management system, and relates to the technical field of medical data management, and the system comprises an adaptive multi-modal data fusion engine, a predictive anomaly tracking module, a distributed knowledge graph reasoning module, a zero-trust safety management module, a man-machine collaborative learning module and a visual decision support module. The self-adaptive multi-modal data fusion engine dynamically integrates multi-source medical examination data through an attention mechanism, and supports real-time updating; the predictive anomaly tracking module analyzes and predicts potential anomalies by using a time sequence and generates intervention suggestions; the distributed knowledge graph reasoning module is combined with a federated learning technology to provide personalized diagnosis support; the zero-trust security management module ensures data security through dynamic verification and block chain evidence storage; the man-machine collaborative learning module optimizes an analysis model based on doctor feedback; the visual decision support module displays the risk distribution in an interactive heat map. According to the invention, the data management efficiency and the diagnosis accuracy can be improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical data management, and more particularly, to an intelligent management system for clinical data of medical tests. Background Art

[0002] With the continuous progress of medical technology and the improvement of the informatization level, the role of medical test data in disease diagnosis, treatment plan formulation, and patient prognosis assessment has become increasingly prominent. Medical test data includes laboratory test results (such as blood routine, biochemical indicators), imaging data (such as X-rays, CT scans), and clinical records (such as doctor observations and patient complaints). These data provide important decision-making bases for doctors. In recent years, the widespread application of electronic medical record systems (EMRs) and laboratory information management systems (LISs) has gradually digitized the collection and storage of medical test data. However, there are still many deficiencies in the existing technology in terms of data management, analysis, and application, which limit its effectiveness in clinical practice.

[0003] First of all, the multi-source data integration ability of existing medical test data management systems is limited. Traditional systems are usually designed for a single data type, for example, only processing laboratory numerical data or imaging data, lacking the ability to uniformly collect and fuse multi-modal data (such as numerical, image, text). Due to the scattered data sources (such as different devices, different departments), manual integration is required, which is not only time-consuming and laborious but also prone to errors. For example, when dealing with the comprehensive data of chronic disease patients, doctors often need to query blood indicators and imaging reports separately and cannot quickly obtain a complete view of the patient's status. In addition, the existing systems lack support for real-time updates of new data, resulting in poor data timeliness and difficulty in meeting the needs of dynamic monitoring.

[0004] Secondly, the degree of intelligence in data analysis and anomaly detection is relatively low. Currently, most systems rely on simple statistical methods (such as threshold judgment) or manual experience to identify abnormal indicators. For example, by setting a fixed range to determine whether blood sugar exceeds the standard. This method is difficult to capture the deep patterns or time trends in the data and cannot predict potential risks. For example, for patients with chronic kidney disease, the existing system may only report the current abnormal blood creatinine level and cannot predict the future deterioration trend based on historical data. In recent years, although some systems have introduced machine learning techniques (such as support vector machines or decision trees), these methods are mostly static analysis, lacking the ability to dynamically model time series data, and the analysis results often lack personalization and clinical relevance.

[0005] Thirdly, there are obvious shortcomings in the existing technology in terms of decision-making support. Traditional systems usually only provide raw data or basic statistical results, and doctors need to interpret and formulate treatment plans by themselves, resulting in low efficiency. Although some advanced systems attempt to integrate expert systems or knowledge bases (such as ICD-10 coding), these systems are mostly based on general rules and are difficult to generate customized suggestions according to the individual characteristics of patients (such as age, medical history). In addition, the existing decision-making support systems lack the ability to adapt, and cannot continuously improve according to doctors' feedback or new cases, resulting in limited accuracy and practicality of the suggestions.

[0006] Finally, data security and privacy protection are a major challenge in the current management of medical test data. With the increasing demand for medical data sharing (such as cross-hospital collaboration, remote consultation), data faces the risk of leakage or tampering during transmission and storage. Most existing systems use basic encryption technologies (such as SSL or RSA), but these methods are difficult to cope with complex network attacks and lack traceability of data operations. For example, when external experts access patient data, existing systems cannot effectively record access logs or dynamically adjust permissions, and it is difficult to meet the strict requirements of regulations such as the General Data Protection Regulation (GDPR).

[0007] In recent years, the application of artificial intelligence (such as deep learning), big data analysis, and blockchain technology in the medical field has gradually emerged, providing new possibilities for solving the above problems. Deep learning technology can process multi-dimensional data and extract complex features, big data analysis can achieve real-time processing of large-scale data, and blockchain provides a means of distributed storage for data security. However, the application of these technologies in the management of medical test data is still in its infancy, and no systematic solution has been formed. Existing research mostly focuses on the optimization of a single technical point, such as only the convolutional neural network (CNN) classification for image data, or only blockchain encryption for data storage, lacking a comprehensive system that integrates multiple technologies and applies them to clinical practice.

[0008] In view of the above problems, this application designs and provides an intelligent management system for clinical data of medical tests to solve these problems. Summary of the Invention

[0009] The purpose of the present invention is to provide an intelligent management system for clinical data of medical tests to solve the problems raised in the above background technology.

[0010] To achieve the above purpose, the present invention provides the following technical solutions:

[0011] The solution of the present invention is an intelligent management system for clinical data of medical tests, which includes an adaptive multi-modal data fusion engine, a predictive anomaly tracking module, a distributed knowledge graph reasoning module, a zero-trust security management module, a human-machine collaborative learning module, and a visualization decision-making support module. The specific functions are as follows:

[0012] Adaptive Multimodal Data Fusion Engine: Collect multi-source medical test data D through standard medical interfaces m , and generate fused data D using the attention mechanism f :

[0013]

[0014] where D m is the data of the m-th modality, M is the number of modalities, w m is the modality weight, W a is the attention weight matrix, b a is the bias;

[0015] Predictive Anomaly Tracking Module: Predict the future anomaly probability P based on the fused data D f : t P

[0016] = σ(W t ·D p + b f ) p )

[0017] where W p is the prediction weight matrix, b p is the bias, and σ is the activation function;

[0018] Distributed Knowledge Graph Inference Module: Combine the anomaly probability P t and the knowledge graph to generate a diagnostic recommendation R:

[0019] R = argmax r∈K (P t ·S r )

[0020] where K is the set of knowledge graph recommendations, and S r is the recommendation matching score;

[0021] Zero-Trust Security Management Module: Encrypt the fused data D f to generate E and record the blockchain hash H b :

[0022] E = AES(D f , K e ), H b = SHA256(E‖T)

[0023] where K e is the encryption key and T is the timestamp;

[0024] Human-Machine Collaborative Learning Module: Update the anomaly probability P according to the doctor's annotation L t:

[0025] P′ t =P t +α(L - P t )

[0026] where α is the learning rate and L is the labeled value; Visual decision support module: Based on the anomaly probability P t generate a risk heat map V i :

[0027]

[0028] where P t (i) is the anomaly probability of the i-th index, μ is the mean, and σ d is the standard deviation; The system realizes the adaptive fusion, predictive analysis, and security management of data through the above modules.

[0029] The solution of the present invention is further optimized as: The adaptive multi-modal data fusion engine calculates the model weight w through the attention mechanism m :

[0030] w m =softmax(W a ·D m +b a )

[0031] where D m is the input modal data, and W a and b a are optimized through training to ensure that the fused data D f reflects the key features.

[0032] The solution of the present invention is further optimized as: The adaptive multi-modal data fusion engine supports incremental updates, and the fusion result D n ′ of the newly added data D f is expressed as:

[0033] D f ′ = D f +β·D n

[0034] where Dn is the newly added modal data, and β is the incremental coefficient, which is determined by the data update frequency.

[0035] The solution of the present invention is further optimized as: The predictive anomaly tracking module generates a hidden state H using a long short-term memory network t :

[0036] H t =LSTM(D f ,H t-1 ,Wp )

[0037] and calculate the anomaly probability P based on Ht t :

[0038] P t =σ(W p ·H t +b p );

[0039] where D f is the fused data, and W p and b p are prediction parameters.

[0040] The solution of the present invention is further optimized as: the predictive anomaly tracking module generates an intervention suggestion I:

[0041] I=f(P t ,C), f(x,y)=if P t >θ then y;

[0042] where Pt is the anomaly probability, C is the patient's condition context, and θ is the threshold.

[0043] The solution of the present invention is further optimized as: the distributed knowledge graph reasoning module updates the graph weight W through federated learning k :

[0044] W′ k =W k +γ·ΔW k

[0045] where W k is the local graph weight, γ is the learning step size, and ΔW k is the external knowledge increment.

[0046] The solution of the present invention is further optimized as: the distributed knowledge graph reasoning module calculates the suggestion matching score Sr:

[0047] S r =cos(P t ,V r )

[0048] where Pt is the anomaly probability, V r is the feature vector of suggestion r, and cos is the cosine similarity.

[0049] The solution of the present invention is further optimized as: the zero-trust security management module generates a trust score T s :

[0050] T s =W t· [U, T, D] + b t

[0051] where T is the timestamp, U is the user identity, D is the device information, W t and b t are verification parameters. The solution of the present invention is further optimized as follows: The abnormal probability P t of the human-machine collaborative learning module is updated as follows:

[0052] P' t = P t + α(L - P t )

[0053] where P t is the initial abnormal probability, L is the doctor's annotation value, and α is the learning rate.

[0054] The solution of the present invention is further optimized as follows: The risk heat map value v i of the visualization decision support module supports user adjustment as follows:

[0055] V' i = V i + δ·w u

[0056] where ν i is the initial heat map value, δ is the adjustment parameter, and w u is the user-defined weight.

[0057] Compared with the prior art, the present invention has the following beneficial effects:

[0058] 1. Through the adaptive multi-modal data fusion engine, the predictive anomaly tracking module, and the distributed knowledge graph reasoning module, the present invention significantly improves the management efficiency and clinical application value of medical test data. The adaptive fusion technology realizes the dynamic integration and real-time update of multi-source data (such as numerical values, images, texts), overcoming the limitations of traditional systems in terms of scattered data and poor timeliness; the predictive anomaly tracking module uses time series analysis to early warn of potential risks, such as predicting abnormal creatinine levels and suggesting interventions, thus shortening the clinical response time; the distributed knowledge graph reasoning module combines global knowledge resources to provide personalized diagnostic suggestions for complex cases, such as rare disease identification, thereby improving the accuracy and comprehensiveness of diagnosis. These functions together enhance the doctor's ability to grasp the patient's condition and optimize the treatment plan formulation process.

[0059] 2. The zero-trust security management module, human-machine collaborative learning module, and visualization decision support module of the present invention further enhance the security and practicality of the system. The zero-trust architecture ensures the security and traceability of data during transmission and sharing through dynamic permission verification and blockchain evidence storage, meeting the requirements of strict privacy regulations; the human-machine collaborative learning module allows doctors to participate in model optimization, making the system's suggestions continuously closer to clinical needs, such as improving the abnormal probability calculation through annotation; the visualization decision support module intuitively displays the risk distribution with an interactive heatmap, helping doctors make quick decisions and explain the condition to patients. These advantages enable the system to be applicable not only to the daily management of hospitals but also to support telemedicine and cross-institutional collaboration, promoting the intelligent and standardized development of medical test data management. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 It is a system block diagram of an intelligent management system for clinical data of medical tests proposed by the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0061] Next, the technical solutions in the embodiments of the present invention will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention. Figure 1

[0062] Figure 1 Referring to

[0063]

[0064] The core of this system consists of six modules, running on high-performance servers (such as computing nodes equipped with NVIDIA A100 GPUs) and distributed databases (such as MongoDB and Neo4j), and implementing functions through a standardized software framework (such as TensorFlow, Hyperledger Fabric). The specific implementation of each module is as follows:

[0064] Adaptive multi-modal data fusion engine: This engine collects multi-source data Dm from inspection devices and electronic medical record systems through HL7 and FHIR interfaces, including numerical (such as blood glucose values), image (such as X-ray films), and text (such as images). Using the attention mechanism to generate fused data Df : where the modal weight w m = softmax(W a ·D m + b a ), W a and b a are initialized by a pre-trained model. When incremental updates are supported, the newly added data Dn is fused into D′ f = D f + β·D n , and β is default set to 0.1.

[0065] Predictive Anomaly Tracking Module: Based on D f , use a long short-term memory network to generate the hidden state H t = LSTM(D f , H t-1 , W p ), and calculate the anomaly probability P t = σ(W p ·H t + b p ), where W p and b p are trained by the patient's historical data. When P t > θ (such as 0.8), combine the disease context (C) to generate the intervention advice I = f(P t , C).

[0066] Distributed Knowledge Graph Reasoning Module: Update the local graph weight W′ k = W k + γ·ΔW k , γ = 0.01. Based on P t , calculate the advice matching score S r = cos(P t , V r ), and generate the diagnostic advice R = argmax r∈K (P t ·S r ).

[0067] Zero-Trust Security Management Module: Encrypt Df as E = AES(D f , K e ), and record the blockchain hash H b = SHA256(E‖T). Access verification is judged by the trust score T s = W t ·[U, T, D]+ b t , and access is rejected when T s < 0.9.

[0068] Human - machine collaborative learning module: After the doctor annotates (L) (e.g., abnormal is 1, normal is 0), update P′ t = P t + α(L - P t ), α = 0.05, to optimize the model weights.

[0069] Visual decision - making support module: Calculate the heat - map value based on Pt Support the user to adjust V′ i = V i + δ·w u , δ and w u are input by the doctor.

[0070] When the system runs, first, the fusion engine collects and processes multi - source data to generate D f ; The prediction module analyzes D f , and outputs P t and I; The knowledge graph module combines P t to generate R; The security module encrypts data and verifies access; The doctor optimizes P through the collaborative module t ; Finally, the visualization module displays the heat - map V i . All operations are completed in real - time on the cloud server, and the database is stored in an encrypted database.

[0071] Example 1: This example shows how the system monitors the serum creatinine level of a chronic kidney disease patient and provides predictive management. The patient is a 45 - year - old male with a 5 - year history of kidney disease and undergoes hemodialysis weekly. The system is deployed on the hospital server, connected to a blood analyzer, an ultrasound device, and an electronic medical record system.

[0072] The adaptive multi - modal data fusion engine is started, and three types of data are collected through the HL7 interface: serum creatinine concentration (numerical type, D1: measured daily in the last week, unit mg / dL, e.g., [2.5, 2.7, 2.9]), kidney ultrasound images (image type, D2: showing kidney atrophy characteristics), and follow - up records (text type, D3: e.g., "The patient's recent edema has worsened"), and the number of modalities M = 3.

[0073] The engine calculates the weights using the attention mechanism. Assume that the pre - trained W a :

[0074] W a = [[0.1, 0.2], [0.3, 0.4], [0.5, 0.6]], b a = [0.1, 0.2], calculate w1 = 0.5, w2 = 0.3, w3 = 0.2, and fuse the data D f = 0.5·D1 + 0.3·D2 + 0.2·D3. The newly added serum creatinine data (2.8 mg / dL) the next day is used as Dn , update D' f = D f + 0.1·2.8.

[0075] The predictive anomaly tracking module receives D' f , and analyzes the time series through LSTM (the hidden layer dimension is set to 64, initial H t -1 = 0), and the trained W p =, b p = [0.05] to calculate H t , and obtain the anomaly probability P for the next 24 hours t = 0.85. Since P t > θ = 0.8, combined with the context C = "Dialysis twice a week", generate the suggestion I = "Increase dialysis to three times a week".

[0076] The distributed knowledge graph reasoning module is based on P t , and updates W through federated learning k (initial W k =, incremental ΔW k is obtained from the regional hospital, γ = 0.01), and calculate S r (such as the renal failure score is 0.9), and obtain R = "The risk of renal failure has increased. It is recommended to adjust the dialysis plan".

[0077] The zero-trust security management module encrypts D f , (K e is a random 256-bit key), generate (E), and record H b on the blockchain to verify the nurse access request

[0078] (U = "Nurse ID123", T = "2025-03-27 09:00", D = "Device IP192.168.1.10"), calculate T s = 0.95 > 0.9, and the authorization passes.

[0079] The human-machine collaborative learning module receives the doctor's annotation L = 1 (confirming the anomaly) and updates P' t = 0.86.

[0080] The visualization decision support module calculates the heat map (μ = 0.5, σ d = 0.2), the doctor adjusts δ = 0.1, w u = 0.2, and updates V'1 = 1.82. Finally, the system warns of abnormal blood creatinine in advance, the doctor adjusts the treatment according to the suggestion, and the patient's edema symptoms are relieved, avoiding acute renal failure.

[0081] Example 2: This example demonstrates the system providing diagnostic support for a chest pain patient in an emergency scenario. The patient is a 60-year-old female who was admitted to the hospital due to sudden chest pain. The emergency room is equipped with a system server connected to an electrocardiograph, a blood analyzer, and an electronic medical record.

[0082] The adaptive multimodal data fusion engine collects data: electrocardiogram (image type, D1: showing ST segment elevation), blood myocardial enzymes (numerical type, D2: troponin I is 0.5 ng / mL, normal < 0.04), and emergency record (text type, D3: such as "chest pain lasted for 30 minutes"), M = 3.

[0083] The attention mechanism calculates w1 = 0.4, w2 = 0.5, w3 = 0.1 (because the weight of myocardial enzymes is the highest), and fuses them into D f The o predictive anomaly tracking module analyzes D f , and the LSTM (hidden layer with 32 dimensions) calculates H t , and obtains P t = 0.92 > θ = 0.8, and it is recommended that I = "Immediately perform electrocardiogram monitoring and prepare for first aid".

[0084] The distributed knowledge graph reasoning module updates W k (ΔW k obtained from the heart disease dataset), calculates S r (myocardial infarction score 0.95), and obtains R = "Suspected acute myocardial infarction, it is recommended to perform coronary angiography". The zero-trust security management module encrypts D f into (E), records H b , and verifies the doctor's access (T s = 0.98), and it passes.

[0085] The human-machine collaborative learning module receives the annotation L = 1 and updates P' t = 0.93. The visualization decision support module generates a heat map (extremely high myocardial enzymes), and the doctor adjusts δ = 0.05, w u = 0.3, and updates V'2 = 9.015. The doctor arranges angiography according to the recommendation, diagnoses myocardial infarction and performs interventional treatment, and the patient is out of danger.

[0086] The content not described in detail in this specification belongs to the prior art well-known to those skilled in the art. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some of the technical features. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. An intelligent management system for clinical data of medical examinations, characterized in that, The system includes an adaptive multi-modal data fusion engine, a predictive anomaly tracking module, a distributed knowledge graph reasoning module, a zero-trust security management module, a human-machine collaborative learning module, and a visualization decision support module. The specific functions are as follows: Adaptive Multimodal Data Fusion Engine: Collect multi-source medical test data D through standard medical interfaces m , and generate fused data D using the attention mechanism f : Among them, D m is the m-th modal data, M is the number of modalities, w m is the modal weight, W a is the attention weight matrix, b a is the bias; Predictive Anomaly Tracking Module: Based on the fused data D f Predict the future anomaly probability P t : P t = σ(W p · D f + b p ) Among them, W p is the prediction weight matrix, b p is the bias, and σ is the activation function; Distributed Knowledge Graph Inference Module: Combining Abnormal Probability P t and Knowledge Graph to Generate Diagnostic Suggestion R: R = argmax r∈K (P t ·S r ) where K is the knowledge graph recommendation set, and S r is the recommendation matching score; Zero-Trust Security Management Module: For the fusion data D f Encrypt and generate E, and record the blockchain hash H b : E = AES(D f , K e ), H b = SHA256(E || T) Among them, K e is the encryption key, and T is the timestamp; Human-computer collaborative learning module: Update the anomaly probability P according to the doctor's annotation L t : P′ t = P t + α(L - P t ) Among them, α is the learning rate, and L is the annotation value; Visual decision support module: Based on the abnormal probability P t Generate a risk heat map V i : Among them, P t (i) is the abnormal probability of the i-th index, μ is the mean value, and σ d is the standard deviation; The system realizes the adaptive fusion, predictive analysis, and security management of data through the above modules.

2. The intelligent management system for clinical data of medical tests according to claim 1, characterized in that, The adaptive multimodal data fusion engine calculates the model weight w through the attention mechanism m : w m = softmax(W a · D m + b a ) Among them, D m is the input modal data, W a and b a Through training and optimization, ensure that the fused data D f reflects the key features.

3. The intelligent management system for clinical data of medical tests according to claim 2, wherein The adaptive multimodal data fusion engine supports incremental updates, and the fusion result D n ' of the newly added data D f ' is expressed as: D f ′ = D f + β·D n Among them, Dn is the newly added modal data, β is the increment coefficient, which is determined by the data update frequency.

4. The intelligent management system for clinical data of medical tests according to claim 1, characterized in that, The predictive anomaly tracking module uses a long short-term memory network to generate a hidden state H t : H t = LSTM(D f , H t-1 , W p ) And calculate the anomaly probability P based on Ht t : P t = σ(W p ·H t + b p ) Among them, D f is the fusion data, W p and b p are the prediction parameters.

5. The intelligent management system for clinical data of medical tests according to claim 4, wherein The predictive anomaly tracking module generates an intervention suggestion I: I = f(P t , C), f(x, y) = if P t > θ then y Among them, Pt is the anomaly probability, C is the patient's condition context, and θ is the threshold.

6. The intelligent management system for clinical data of medical examinations according to claim 1, wherein The distributed knowledge graph reasoning module updates the graph weight W through federated learning k : W′ k = W k + γ·ΔW k Among them, W k is the local atlas weight, γ is the learning step size, and ΔW k is the external knowledge increment.

7. The intelligent management system for clinical data of medical tests according to claim 6, characterized in that, The distributed knowledge graph reasoning module calculates the recommendation matching score Sr: S r = cos(P t , V r ) Among them, Pt is the abnormal probability, V r is the eigenvector of the recommended r, and cos is the cosine similarity.

8. The intelligent management system for clinical data of medical tests according to claim 1, characterized in that The zero-trust security management module generates a trust score T s : T s = W t · [U, T, D] + b t Among them, T is the timestamp, U is the user identity, D is the device information, and W t and b t are verification parameters.

9. The intelligent management system for clinical data of medical tests according to claim 1, characterized in that The abnormal update probability P of the human-machine collaborative learning module t : P′ t = P t + α(L - P t ) Among them, P t is the initial abnormal probability, L is the doctor's marked value, and α is the learning rate.

10. A medical test clinical data intelligent management system according to claim 1, characterized in that, The risk heat map value v of the visual decision support module i Supports user adjustment: V′ i = V i + δ·w u Among them, ν i is the initial heat map value, δ is the adjustment parameter, and w u is the user-defined weight.

Images