Brain lesion detection method and system based on multi-modal data

Abstract

The invention discloses a brain lesion detection method and system based on multi-modal data, and relates to the technical field of brain disease detection. The invention discloses a brain lesion detection system based on multi-modal data. The brain lesion detection system comprises a brain lesion detection triggering module, a brain lesion detection primary screening module and a brain lesion detection secondary screening module. According to the method, screening is performed in combination with current and historical treatment records, so that potential brain lesion clues can be found earlier, and the possibility of early diagnosis is improved; a brain variation feature extraction model is combined with a disease-medication-parameter element path and three attention channel feature extraction method, so that lesion features can be analyzed more comprehensively and deeply, and the accuracy of preliminary screening is improved; a multi-stage elimination mechanism is introduced through a brain transformer accurate re-screening model, and parallel mixed feature analysis and time-space data are combined, so that misjudgment can be effectively reduced, and the accuracy of final diagnosis is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of brain disease detection, and in particular to a brain lesion detection method and system based on multimodal data. Background Art

[0002] Current brain lesions, such as Alzheimer's disease, Parkinson's disease, and stroke, develop insidiously and are difficult to detect in the early stages. They also have a high incidence rate and are prone to complications. However, existing detection methods have limitations: traditional MRI cannot accurately detect tiny neurodegenerative lesions, and when the lesions cause obvious symptoms, they have often entered an irreversible stage, reducing the effectiveness of treatment. At the same time, many patients also suffer from chronic diseases, but the potential association between these physiological data and brain lesions has not been fully explored and utilized. In addition, existing detection methods lack cross-system, multi-modal linkage screening capabilities, making it difficult for high-risk groups to obtain effective intervention in a timely manner. Since additional examinations are usually cumbersome and costly, many patients choose to give up early screening, further reducing the possibility of early intervention.

[0003] Therefore, there is an urgent need for a non-invasive detection method that reduces additional medical examinations, improves the early detection rate of brain lesions through correlation analysis and feature extraction of multimodal data, and provides accurate detection and prediction solutions. Summary of the invention

[0004] The present invention aims to provide a brain lesion detection method and system based on multimodal data to improve the accuracy of early brain lesion diagnosis.

[0005] A brain lesion detection method based on multimodal data comprises the following steps:

[0006] When receiving the current medical records of patients with potential brain changes, screen the current medical records and historical medical records to obtain clues of potential brain lesions; perform feature analysis on the clues of potential brain lesions to obtain initial screening labels for brain changes; patients with potential brain changes have at least one potential brain change complication and meet the common brain change conditions;

[0007] The brain change feature extraction model is used to generate clues of potential brain lesions, and the clues of potential brain lesions are processed to generate preliminary screening labels of brain changes. The brain change feature extraction model is used to combine the disease-medication-parameter meta-path and the three-attention channel feature extraction method to predict the spread of brain changes and generate preliminary screening labels of brain changes;

[0008] Potential brain degeneration patients whose initial screening labels meet the characteristic indicators of brain degeneration are marked as brain degeneration patients to be screened;

[0009] Perform secondary brain lesion detection on patients with screened brain changes to obtain the results of brain change re-screening; use the brain change precise re-screening model to perform secondary brain lesion detection, introduce a multi-level elimination mechanism and judge and screen through parallel hybrid features, and combine spatio-temporal data to improve the accuracy of the brain change re-screening results.

[0010] As a preferred technical solution of the present invention, the specific steps for screening the current medical record and historical medical records include:

[0011] The brain change feature extraction model includes a data linkage layer, a clue extraction layer, and a label output layer;

[0012] The data linkage layer is used to extract the big data of the symptoms of potential brain change complications corresponding to potential brain change patients to obtain brain change induction data; screen the brain change induction data and the current medical record and historical medical records to obtain the brain change induction multimodal data D n , n = 1, 2,..., N; N is the total number of dimensions of brain change induction factors;

[0013] The clue extraction layer is used to perform feature analysis based on the brain change induction multimodal data D n to obtain potential brain lesion clues;

[0014] The label output layer is used to perform feature output based on the potential brain lesion clues to obtain the initial brain change screening label.

[0015] As a preferred technical solution of the present invention, the specific steps for performing feature analysis in the clue extraction layer include:

[0016] For the brain change induction multimodal data D n generate the disease-symptom-medication-parameter meta-path to obtain the brain change clue extraction meta-path L n ;

[0017] Construct three attention channels in the clue extraction layer. The three attention channels are pathological feature attention, treatment response attention, and social behavior attention, and the internal structure is a parallel decoupling structure; input the brain change clue extraction meta-path L n into the three attention channels for parallel analysis to obtain the brain change induction feature Z n ;

[0018] At the same time, extract the image data in all the brain change induction multimodal data D n to obtain the brain change meta-image data;

[0019] Map the brain change meta-image data to a density matrix in the Hilbert space to obtain the brain change analysis matrix;

[0020] For the brain change analysis matrix and the brain change induction feature Z nComprehensive clue extraction is carried out using cross-modal superposition quantum gates to obtain potential brain lesion clues.

[0021] As a preferred technical solution of the present invention, the specific steps for feature output in the label output layer include:

[0022] Using the potential brain lesion clues to generate a prediction of the spread result of brain lesions, obtaining a prediction result of brain lesion spread;

[0023] Performing quantization label conversion on the prediction result of brain lesion spread to obtain a preliminary screening label for brain lesions;

[0024] Among them, using the formula Calculate to obtain the prediction result of brain lesion spread; X represents potential brain lesion clues, t represents the time axis of brain lesion development, W represents the prediction coefficient of brain lesion spread, ΔX represents the result value calculated by using the Laplace operator on potential brain lesion clues, and f(X) represents the non-linear brain lesion reflection term.

[0025] As a preferred technical solution of the present invention, the specific steps for secondary brain lesion detection of patients to be screened for brain lesions include:

[0026] Obtain the special examination data of brain lesions of the patients to be screened for brain lesions;

[0027] Input the special examination data of brain lesions into the accurate re-screening model for brain lesions for diagnosis. The accurate re-screening model for brain lesions includes a multi-level elimination mechanism layer and a brain lesion detection layer;

[0028] Perform parallel hybrid feature judgment on several brain pathological features in the special examination data of brain lesions. If no less than m brain pathological features meet the brain lesion screening standard, key images of brain lesions are collected for the patients to be screened for brain lesions, obtaining key images of brain lesions;

[0029] Combining the special examination data of brain lesions and the key images of brain lesions, use a quantum annealing machine to solve the optimal classification hyperplane to obtain a multi-dimensional brain lesion risk vector R, R = (R 1 ,R 2 ,…,R I ), where I is the vector dimension in the multi-dimensional brain lesion risk vector R;

[0030] Input the multi-dimensional brain lesion risk vector R into the brain lesion detection layer to obtain the re-screening result of brain lesions.

[0031] As a preferred technical solution of the present invention, the specific steps for obtaining the re-screening result of brain lesions in the brain lesion detection layer include:

[0032] Extract the prediction result of brain lesion spread of the patients to be screened for brain lesions to obtain the preliminary screening prediction result of brain lesion spread;

[0033] Align the preliminary screening results of brain lesion diffusion prediction and the multi-dimensional brain lesion risk vector R in space-time to obtain brain lesion risk diagnosis data;

[0034] Encode the brain lesion knowledge graph into a brain lesion attention bias matrix and introduce it into dynamic convolutional attention to perform brain lesion detection on the brain lesion risk diagnosis data, and obtain the final brain lesion re-screening results.

[0035] A brain lesion detection system based on multi-modal data, comprising:

[0036] A brain lesion detection trigger module, including a data screening unit; the data screening unit is used to screen the current medical record and historical medical records when receiving the current medical record of a potential brain lesion patient, so as to obtain potential brain lesion clues; perform feature analysis on the potential brain lesion clues to obtain a preliminary brain lesion screening label; the potential brain lesion patient has at least one potential brain lesion complication and meets the common brain lesion conditions;

[0037] A preliminary brain lesion detection module, including an anomaly detection unit; the anomaly detection unit is used to generate potential brain lesion clues by using a brain lesion feature extraction model, and process the potential brain lesion clues to generate a preliminary brain lesion screening label. The brain lesion feature extraction model is used to combine the disease-medication-parameter meta-path and the three-attention channel feature extraction method to perform brain lesion diffusion prediction and generate a preliminary brain lesion screening label; for potential brain lesion patients whose preliminary brain lesion screening labels meet the brain lesion feature indicators, they are marked as brain lesion patients to be screened;

[0038] A brain lesion detection re-screening module, including a precise detection unit; the precise detection unit is used to perform secondary brain lesion detection on the brain lesion patients to be screened to obtain brain lesion re-screening results; use a brain lesion precise re-screening model to perform secondary brain lesion detection, introduce a multi-level elimination mechanism and make judgments and screenings through parallel hybrid features, and combine space-time data to improve the accuracy of the brain lesion re-screening results.

[0039] The present invention has the following advantages:

[0040] 1. By combining the current and historical medical records for screening, the present invention can detect potential brain lesion clues earlier and improve the possibility of early diagnosis; by using the brain lesion feature extraction model to combine the disease-medication-parameter meta-path and the three-attention channel feature extraction method, it can analyze the lesion features more comprehensively and deeply, and improve the accuracy of the preliminary screening; by considering the potential complications that the patient may have and the common brain lesion conditions, it can analyze the lesion risks from multiple angles and improve the reliability of the screening; by introducing a multi-level elimination mechanism into the brain lesion precise re-screening model and combining parallel hybrid feature analysis and space-time data, it can effectively reduce misjudgments and improve the accuracy of the final diagnosis.

[0041] 2. The present invention establishes the association among the pathological features, treatment history, and physiological parameters of patients through the disease-medication-parameter meta-path, enabling the screening to be not limited to a single factor but to be analyzed based on complete medical data. By utilizing multi-source information such as imaging data, pathological data, and social behavior data, a more complete patient health portrait is constructed to improve the accuracy of extracting clues for brain lesions. The pathological feature attention is used to focus on the biological features of the disease, the treatment response attention is used to analyze the patient's medication history and treatment effects, and the social behavior attention is used to comprehensively consider the impact of the patient on cognitive and neurological functions. The introduction of the parallel decoupling structure can avoid information redundancy, ensuring that each attention channel focuses on features in different dimensions and improving the interpretability and adaptability of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 FIG. is a schematic structural diagram of a brain lesion detection system based on multi-modal data adopted in an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] In order to enable those skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0044] Embodiment 1. A brain lesion detection method based on multi-modal data includes the following steps:

[0045] When receiving the current medical record of a potential brain lesion patient, screen the current medical record and historical medical records to obtain potential brain lesion clues; perform feature analysis on the potential brain lesion clues to obtain a preliminary brain lesion screening label; the potential brain lesion patient has at least one potential brain lesion complication and meets the common brain lesion conditions.

[0046] For example, the groups meeting the conditions of potential brain lesion patients mainly include the following types of people: those with a history of brain diseases, long-term chronic disease patients, patients who have long-term used nervous system drugs, or patients with high-risk living habits or environmental factors; potential brain lesion complications refer to complications that may cause or exacerbate brain lesions, such as cerebrovascular-related complications, nervous system diseases, metabolic and systemic diseases, such as hypertension, diabetes, chronic renal insufficiency, hypothyroidism, etc.; the common brain lesion conditions specifically include imaging abnormalities, cognitive function decline, motor function abnormalities, neuropsychiatric symptoms, older age, etc., which will all make a patient a potential brain lesion patient.

[0047] By screening current and historical medical records, potential clues of brain lesions can be detected earlier, preventing patients from missing the best intervention opportunity due to the lack of obvious initial symptoms of the lesions. Traditional brain lesion screening relies on doctors' experience and manual analysis, and often, relatively severe brain lesion symptoms have already appeared. However, this method uses systematic automatic screening and feature analysis, significantly reducing the time cost of manual screening and improving the utilization efficiency of medical resources. By comprehensively analyzing medical history, medication use, and relevant parameters, further tests are only conducted on patients who meet the conditions of brain lesion complications and common brain lesions, avoiding unnecessary screening and improving the accuracy of screening.

[0048] Use the brain lesion feature extraction model to generate potential clues of brain lesions and process the potential clues of brain lesions to generate preliminary screening labels for brain lesions. The brain lesion feature extraction model is used to combine the disease-medication-parameter meta-path and the three-attention channel feature extraction method for brain lesion spread prediction to generate preliminary screening labels for brain lesions.

[0049] The specific steps for screening current and historical medical records include:

[0050] The brain lesion feature extraction model includes a data linkage layer, a clue extraction layer, and a label output layer.

[0051] The data linkage layer is used to extract the big data of the symptoms of potential brain lesion complications corresponding to potential brain lesion patients to obtain brain lesion induction data; screen the brain lesion induction data and current and historical medical records to obtain brain lesion induction multimodal data D n , n = 1, 2,..., N; N is the total number of dimensions of brain lesion induction factors;

[0052] The clue extraction layer is used to perform feature analysis based on the brain lesion induction multimodal data D n to obtain potential clues of brain lesions;

[0053] The label output layer is used to perform feature output based on the potential clues of brain lesions to obtain preliminary screening labels for brain lesions.

[0054] The data linkage layer integrates multimodal data such as the medical records, medical history, and medication use of potential brain lesion patients, which helps to discover hidden clues of brain lesions and improve the early detection rate of brain lesions; through big data analysis, it is possible to extract symptom information related to potential brain lesion complications, avoid ignoring potential risks due to individual factors, and improve the accuracy of screening.

[0055] Through the brain lesion induction multimodal data, fully consider multi-dimensional factors, avoid misjudgment that may be caused by relying solely on a single indicator, and improve the comprehensiveness of disease recognition; combined with the three-attention channel feature extraction, it can highlight key features, improve the sensitivity of the model to the risk of brain lesions, and reduce the risks of misdiagnosis and missed diagnosis.

[0056] The clue extraction layer uses feature analysis technology to mine valuable potential lesion clues from multi-modal data. Automated analysis can quickly locate high-risk patients, improve the screening efficiency of hospitals, and reduce the workload of doctors. The label output layer intelligently outputs the extracted brain lesion features to generate preliminary brain lesion screening labels, providing a clear basis for further accurate detection. Through the labeled screening results, risk grading can be achieved, providing a more intuitive diagnostic reference for doctors, optimizing the screening process, and improving the rational allocation of medical resources.

[0057] For example, diabetes is a common potential complication of brain lesions. Its long-term development may lead to cerebrovascular lesions, cognitive decline, and neurodegenerative diseases. If diabetes is analyzed as a potential complication of brain lesions, the symptom big data may include the following aspects: the association between blood glucose fluctuations and cognitive decline, the association between diabetes and cerebral small vessel lesions, the increased risk of hippocampal atrophy in diabetic patients, the neuroinflammatory response caused by diabetes, the correlation between the frequency of hypoglycemic attacks and Parkinson's syndrome-like symptoms, etc. These clues help screen high-risk diabetic patients and determine whether they are likely to develop brain lesions.

[0058] The specific steps for feature analysis in the clue extraction layer include:

[0059] For the brain lesion-induced multi-modal data D n Generate the disease-medication-parameter meta-path to obtain the brain lesion clue extraction meta-path L n ;

[0060] Construct three attention channels in the clue extraction layer. The three attention channels are pathological feature attention, treatment response attention, and social behavior attention, and the internal structure is a parallel decoupling structure. Input the brain lesion clue extraction meta-path L n into the three attention channels for parallel analysis to obtain the brain lesion-induced feature Z n ;

[0061] Meanwhile, extract the imaging data from all brain lesion-induced multi-modal data D n to obtain the brain lesion meta-imaging data.

[0062] Map the brain lesion meta-imaging data to a density matrix in the Hilbert space to obtain the brain lesion analysis matrix.

[0063] Use cross-modal superposition quantum gates to comprehensively extract clues from the brain lesion analysis matrix and the brain lesion-induced feature Z n to obtain potential brain lesion clues.

[0064] By establishing the association between the pathological features, treatment history, and physiological parameters of patients through the disease-medication-parameter meta-path, the screening is not limited to a single factor but is analyzed based on complete medical data. By using multi-source information such as imaging data, pathological data, and social behavior data, a more complete patient health profile is constructed to improve the accuracy of extracting clues for brain lesions. Pathological feature attention is used to focus on the biological features of the disease, such as imaging manifestations, clinical symptoms, and disease progression trends. Treatment response attention is used to analyze the patient's medication history and treatment effects to determine whether the treatment affects brain health. Social behavior attention is used to comprehensively consider external factors such as the patient's lifestyle, sleep pattern, and social interaction and their impact on cognitive and neurological functions. The introduction of a parallel decoupling structure can avoid information redundancy, ensure that each attention channel focuses on features in different dimensions, and improve the interpretability and adaptability of the model.

[0065] Extract the imaging data from all brain lesion-induced multimodal data and convert it into a density matrix in Hilbert space to make the data have better feature expression ability mathematically. The quantization mapping of imaging data helps to improve the accuracy of image recognition and the sensitivity of the screening system to subtle lesions such as white matter lesions and brain atrophy. Through cross-modal superposition quantum gates, deep fusion of imaging data and disease-medication-parameter data is achieved, reducing the deviation between cross-modal data and improving the complementarity of the data. The quantum computing method makes data mapping more efficient, can calculate potential brain lesion clues faster, and improves the computational efficiency of screening.

[0066] The specific steps for feature output in the label output layer include:

[0067] Use the potential brain lesion clues to predict the generation of brain lesion diffusion results and obtain the brain lesion diffusion prediction results.

[0068] Perform quantization label conversion on the brain lesion diffusion prediction results to obtain the preliminary brain lesion screening labels.

[0069] Among them, the brain lesion diffusion prediction results are calculated using the formula X represents the potential brain lesion clues, t represents the time axis of brain lesion development, W represents the predicted brain lesion diffusion coefficient, which is set by professional technicians according to the actual situation. ΔX represents the result value obtained by calculating the potential brain lesion clues using the Laplace operator, and f(X) represents the non-linear brain lesion response term.

[0070] Calculate the brain lesion diffusion prediction results through formulas, and use the Laplace operator to calculate the potential brain lesion clues, which can quantify the diffusion trend of brain lesions and improve the prediction ability of disease development; calculate the brain lesion diffusion prediction results in combination with the time axis, enabling the screening system to dynamically monitor the disease progression instead of static analysis, and improving the timeliness of detection; convert the prediction results into preliminary brain lesion screening labels to achieve stratified screening of patients, facilitating doctors to quickly identify high-risk patients and optimize medical decisions; the non-linear brain lesion reflection term is used to capture the complex dynamic changes during the brain lesion process, and its training mainly involves data preprocessing, feature extraction, mathematical modeling, parameter optimization and verification;

[0071] For potential brain lesion patients whose preliminary brain lesion screening labels meet the brain lesion characteristic indicators, they are labeled as patients to be screened for brain lesions;

[0072] The brain lesion characteristic indicators are set by professional technical personnel according to the actual situation; professional technical personnel can set the screening threshold according to specific needs. The brain lesion characteristic indicators should combine multiple factors such as imaging, cognitive function, blood biomarkers, medical history, physical signs, and lifestyle to ensure the accuracy and comprehensiveness of screening and improve the ability to early identify potential brain lesion patients;

[0073] Conduct secondary brain lesion detection on patients to be screened for brain lesions to obtain the results of secondary brain lesion screening; use the precise secondary brain lesion screening model for secondary brain lesion detection, introduce a multi-level elimination mechanism and judge and screen through parallel mixed features, and combine spatio-temporal data to improve the accuracy of the secondary brain lesion screening results;

[0074] The specific steps for conducting secondary brain lesion detection on patients to be screened for brain lesions include:

[0075] Obtain the special brain lesion examination data of patients to be screened for brain lesions; among them, the patients to be screened for brain lesions are screened by the screening center or hospital information system according to the preliminary brain lesion screening results, give the reasons for the need for further special examinations, and conduct further screening according to the patient's wishes to obtain the special brain lesion examination data;

[0076] Input the special brain lesion examination data into the precise secondary brain lesion screening model for diagnosis. The precise secondary brain lesion screening model includes a multi-level elimination mechanism layer and a brain lesion detection layer;

[0077] Perform parallel mixed feature judgment on several brain pathological features in the special brain lesion examination data. If no less than m brain pathological features meet the brain lesion screening criteria, key images of the brain lesions will be collected for patients to be screened for brain lesions to obtain key brain lesion images;

[0078] Among them, for patients with less than m brain pathological features meeting the screening criteria for brain lesions, that is, individuals who do not meet the clear screening criteria for brain lesions but may have certain risks, hospitals and medical institutions can adopt hierarchical management or personalized intervention measures;

[0079] Combining the special inspection data of brain lesions and the key images of brain lesions, using a quantum annealing machine to solve the optimal classification hyperplane, a multi-dimensional brain lesion risk vector R is obtained, R=(R 1 , R 2 , …, R I ), where I is the vector dimension in the multi-dimensional brain lesion risk vector R;

[0080] Input the multi-dimensional brain lesion risk vector R into the brain lesion detection layer to obtain the re-screening result of brain lesions;

[0081] By obtaining the special inspection data of brain lesions and conducting secondary analysis, the accuracy of detection can be improved and misjudgment can be reduced; adopting a multi-level elimination mechanism to ensure that only truly high-risk patients enter the next screening step, avoiding waste of medical resources and optimizing the screening process; in the special inspection data of brain lesions, if not less than m brain lesion characteristics are met, further key image acquisition is carried out to reduce the situation of misjudgment of single characteristics; dynamically adjusting the weights of different characteristics, comprehensively analyzing based on pathological, imaging, and cognitive data, to improve the adaptability of the model to different types of brain lesions; after preliminary feature screening, key image acquisition such as MRI, PET-CT, etc. is carried out to reduce unnecessary imaging examinations, improve the utilization efficiency of medical resources, and save medical expenses for patients;

[0082] By using a quantum annealing machine to solve the optimal classification hyperplane, the optimal risk assessment boundary can be quickly calculated under multi-dimensional data, improving the calculation efficiency of screening; generating a multi-dimensional brain lesion risk vector can more finely evaluate different types of brain lesion risks;

[0083] The specific steps to obtain the re-screening result of brain lesions in the brain lesion detection layer include:

[0084] Extract the brain lesion diffusion prediction result of the patient to be screened for brain lesions to obtain the preliminary screening brain lesion diffusion prediction result;

[0085] Align the preliminary screening brain lesion diffusion prediction result and the multi-dimensional brain lesion risk vector R in space and time to obtain the brain lesion risk diagnosis data;

[0086] Use the brain lesion knowledge graph to encode it into a brain lesion attention bias matrix and introduce it into the dynamic convolution attention to perform brain lesion detection on the brain lesion risk diagnosis data to obtain the final re-screening result of brain lesions;

[0087] The brain lesion detection layer realizes more accurate, efficient, and intelligent re-screening detection of brain lesions through advanced technologies such as brain lesion diffusion prediction, multi-dimensional risk vector alignment, knowledge graph encoding, and dynamic convolutional attention mechanism. The preliminary screening results provide early lesion trend prediction based on the lesion diffusion trend, and the multi-dimensional brain lesion risk vectors provide more detailed individual risk scores. Combining multi-modal data such as images, biomarkers, and cognitive tests enhances the prediction accuracy. Based on a large-scale medical database containing clinical cases, medical research, molecular biology information, etc., a causal relationship map of brain lesions is established. Through knowledge graph reasoning of relevance, high-risk factor combinations are automatically identified. For example, certain specific biomarkers plus imaging abnormalities may be highly correlated with specific types of neurodegenerative diseases. Matching with the patient's individual data improves the accuracy of disease classification and reduces misdiagnosis and missed diagnosis. The weights of traditional neural networks are fixed and may not be able to adapt to the brain lesion characteristics of different individuals. Dynamic convolutional attention can adjust the feature weights for different patients to ensure that the most relevant lesion features receive higher attention. For example, for Alzheimer's disease, the model may pay more attention to hippocampal atrophy and β-amyloid levels, while for stroke, it may pay more attention to white matter lesions and vascular factors. The attention mechanism filters out irrelevant information and reduces misjudgment caused by factors such as data errors and imaging artifacts.

[0088] During the brain lesion detection process of the present invention, intelligent screening through early calculation can effectively reduce unnecessary high-cost examinations in the follow-up, optimize the allocation of medical resources. At the same time, misdiagnosis will lead to patients receiving unnecessary treatments, increasing medical costs; missed diagnosis will delay the best intervention time, resulting in the aggravation of the patient's condition, and the later treatment cost far exceeds the early calculation cost. Moreover, early calculation helps quickly lock in high-risk patients, can optimize hospital resources, improve the screening efficiency, reduce the misdiagnosis rate, reduce the high treatment cost of patients in the later stage, and improve the overall operation efficiency of the hospital and the patient health management level.

[0089] Example 2, a brain lesion detection system based on multi-modal data, as shown in Figure 1 shown, includes:

[0090] The brain lesion detection trigger module includes a data screening unit. The data screening unit is used to screen the current medical record and historical medical records when receiving the current medical record of a potential brain lesion patient to obtain potential brain lesion clues. Feature analysis is performed on the potential brain lesion clues to obtain a preliminary brain lesion screening label. The potential brain lesion patient has at least one potential brain lesion complication and meets the common brain lesion conditions.

[0091] The brain lesion detection preliminary screening module includes an anomaly detection unit; the anomaly detection unit is used to generate potential brain lesion clues by using the brain lesion feature extraction model, and process the potential brain lesion clues to generate brain lesion preliminary screening labels. The brain lesion feature extraction model is used to combine the disease-medication-parameter meta-path and the three-attention-channel feature extraction method to predict brain lesion diffusion and generate brain lesion preliminary screening labels; for potential brain lesion patients whose brain lesion preliminary screening labels meet the brain lesion feature indicators, they are marked as patients to be screened for brain lesions.

[0092] The brain lesion detection re-screening module includes a precise detection unit; the precise detection unit is used to perform secondary brain lesion detection on patients to be screened for brain lesions to obtain the brain lesion re-screening results; use the brain lesion precise re-screening model to perform secondary brain lesion detection, introduce a multi-level elimination mechanism and make judgments and selections through parallel hybrid features, and combine spatio-temporal data to improve the accuracy of the brain lesion re-screening results.

[0093] It should be understood that those of ordinary skill in the art can make improvements or transformations according to the above description, and all such improvements and transformations should fall within the protection scope of the appended claims of the present invention. The parts not described in detail in this specification belong to the prior art well-known to those skilled in the art.

Claims

1. A brain lesion detection method based on multimodal data, characterized in that: The following steps are involved: When receiving the current medical records of patients with potential brain changes, the current medical records and historical medical records are screened to obtain clues of potential brain lesions; characteristic analysis is performed on the clues of potential brain lesions to obtain initial screening labels for brain changes; Patients with potential brain degeneration had at least one potential brain degeneration complication and met the conditions for common brain degeneration; The brain change feature extraction model is used to generate clues of potential brain lesions, and the clues of potential brain lesions are processed to generate preliminary screening labels of brain changes. The brain change feature extraction model is used to combine the disease-medication-parameter meta-path and the three-attention channel feature extraction method to predict the spread of brain changes and generate preliminary screening labels of brain changes; Potential brain degeneration patients whose initial screening labels meet the characteristic indicators of brain degeneration are marked as brain degeneration patients to be screened; Conduct a second brain lesion test on patients to be screened for brain changes to obtain the results of the brain change re-screening; A precise re-screening model for brain changes is used for secondary brain lesion detection. A multi-level elimination mechanism is introduced and judgment and screening are performed through parallel mixed features. The accuracy of brain change re-screening results is improved by combining spatiotemporal data.

2. The method for detecting brain lesions based on multimodal data according to claim 1, characterized in that: Specific steps for screening current and historical medical records include: The brain change feature extraction model includes a data linkage layer, a clue extraction layer, and a label output layer; The data linkage layer is used to extract the big data of symptoms of potential brain degeneration patients corresponding to potential brain degeneration complications to obtain brain degeneration induced data; the brain degeneration induced data and the current medical records and historical medical records are screened to obtain brain degeneration induced multimodal data D n , n = 1, 2, ..., N; N is the total number of dimensions of brain change inducing factors; The clue extraction layer is used to induce multimodal data D according to brain changes n Conduct feature analysis to obtain clues to potential brain lesions; The label output layer is used to output features based on potential brain lesion clues to obtain preliminary screening labels for brain changes.

3. The method for detecting brain lesions based on multimodal data according to claim 2, characterized in that: The specific steps of feature analysis in the clue extraction layer include: For brain change induced multimodal data D n Generate the disease-drug-parameter metapath to obtain the brain change clue extraction metapath L n ; In the clue extraction layer, three attention channels are constructed. The three attention channels are pathological feature attention, treatment response attention, and social behavior attention. The internal structure is a parallel decoupling structure. The brain change clue extraction meta-path L n Input to the three attention channels for parallel analysis to obtain the brain change induced feature Z n ; At the same time, all brain change-induced multimodal data D n The image data in the image data are obtained to obtain the brain variable image data; Mapping brain variable image data into a density matrix in Hilbert space to obtain a brain variable analysis matrix; Brain change analysis matrix and brain change induced feature Z n Cross-modal superposition quantum gates are used to perform comprehensive clue extraction to obtain clues to potential brain lesions.

4. The method for detecting brain lesions based on multimodal data according to claim 3, characterized in that: The specific steps for feature output in the label output layer include: Using potential brain lesion clues to predict brain change spread results, the prediction results of brain change spread are obtained; Perform quantitative label conversion on the prediction results of brain degeneration diffusion to obtain the initial screening labels of brain degeneration; Among them, using the formula The prediction results of brain change diffusion are obtained by calculation; X represents the potential brain lesion clues, t represents the time axis of brain change development, W represents the predicted brain change diffusion coefficient, ΔX represents the result value of calculating the potential brain lesion clues using the Laplace operator, and f(X) represents the nonlinear brain change reflection term.

5. The method for detecting brain lesions based on multimodal data according to claim 4, characterized in that: Specific steps for secondary brain lesion testing in patients with ethmoidal encephalopathy include: Obtain the brain degeneration-specific examination data of patients to be screened for brain degeneration; The brain change special examination data is input into the brain change precision re-screening model for diagnosis. The brain change precision re-screening model includes a multi-level elimination mechanism layer and a brain change detection layer; Perform parallel mixed feature judgment on several brain pathological features in the brain degeneration special examination data. If no less than m brain pathological features meet the brain lesion screening criteria, key images are collected for the patients to be screened for brain degeneration to obtain key images of brain degeneration; Combining the special examination data of brain degeneration and the key images of brain degeneration, the optimal classification hyperplane is solved by using a quantum annealing machine to obtain a multidimensional brain degeneration risk vector R, R = (R1, R2, ..., R I ), I is the vector dimension in the multidimensional brain degeneration risk vector R; The multidimensional brain degeneration risk vector R is input into the brain degeneration detection layer to obtain the brain degeneration re-screening result.

6. The method for detecting brain lesions based on multimodal data according to claim 5, characterized in that: The specific steps for obtaining the brain change re-screening results in the brain change detection layer include: Extract the brain degeneration diffusion prediction results of the patients to be screened for brain degeneration, and obtain the initial screening brain degeneration diffusion prediction results; The prediction results of brain change diffusion in the initial screening and the multidimensional brain change risk vector R are aligned in time and space to obtain the brain change risk diagnosis data. The brain change knowledge graph is encoded into a brain change attention bias matrix, which is introduced into the dynamic convolutional attention to perform brain change detection on the brain change risk diagnosis data to obtain the final brain change re-screening results.

7. A brain lesion detection system based on multimodal data, characterized in that: The system applies a brain lesion detection method based on multimodal data as described in any one of claims 1 to 6, comprising: The brain change detection trigger module includes a data screening unit; the data screening unit is used to receive the current medical records of the potential brain change patient, screen the current medical records and historical medical records, and obtain clues of potential brain lesions; perform feature analysis on the clues of potential brain lesions to obtain a preliminary screening label of brain change; the potential brain change patient has at least one potential brain change complication and meets the common brain change conditions; The brain change detection and screening module includes an anomaly detection unit; the anomaly detection unit is used to generate potential brain lesion clues using a brain change feature extraction model, and process the potential brain lesion clues to generate brain change screening labels. The brain change feature extraction model is used to combine the symptom-drug-parameter metapath and the three-attention channel feature extraction method to predict the spread of brain changes and generate brain change screening labels; potential brain change patients whose brain change screening labels meet the brain change feature indicators are marked as brain change patients to be screened; The brain change detection rescreening module includes a precise detection unit; the precise detection unit is used to perform secondary brain lesion detection on patients to be screened for brain changes to obtain brain change rescreening results; a brain change precise rescreening model is used to perform secondary brain lesion detection, a multi-level elimination mechanism is introduced, and judgment and screening are performed through parallel mixed features, and spatiotemporal data are combined to improve the accuracy of brain change rescreening results.

Images