Cardiovascular lesion parameter identification method and system based on data analysis

Abstract

The invention relates to the technical field of cardiovascular disease diagnosis, and discloses a cardiovascular disease parameter identification method and system based on data analysis, and the method comprises the steps: collecting cardiovascular image data, physiological parameter data and clinical data of a patient, obtaining multi-source data, carrying out the noise reduction and missing value and abnormal value processing of the multi-source data, and obtaining preprocessed data; calculating a mutual information value between each feature and the cardiovascular lesion by adopting a feature selection algorithm based on mutual information to obtain a target feature; encoding the target features, inputting the encoded target features into a parameter identification model, and processing the encoded target features by the parameter identification model to obtain a cardiovascular lesion parameter identification result; according to the invention, accurate and rapid identification of cardiovascular disease parameters is realized, and the accuracy and efficiency of cardiovascular disease diagnosis are improved.

Description

Technical Field

[0001] This invention relates to the field of cardiovascular disease diagnosis technology, and specifically to a method and system for identifying cardiovascular lesion parameters based on data analysis. Background Technology

[0002] Cardiovascular disease is one of the major diseases that endanger human health. Accurate identification of cardiovascular lesion parameters is of vital importance for the early diagnosis, treatment planning and prognosis assessment of cardiovascular diseases. Traditional methods of identifying cardiovascular lesion parameters, such as manual interpretation by doctors through medical images (such as CT, MRI, angiography, etc.), are not only inefficient, but also easily affected by the doctor's subjective factors, resulting in poor accuracy and consistency of diagnostic results. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing a method and system for identifying cardiovascular lesion parameters based on data analysis.

[0004] The first aspect of this invention provides a method for identifying cardiovascular lesion parameters based on data analysis, which involves collecting cardiovascular imaging data, physiological parameter data, and clinical data from patients to obtain multi-source data, and performing noise reduction, missing value and outlier processing on the multi-source data to obtain preprocessed data; A feature selection algorithm based on mutual information is used to calculate the mutual information value between each feature and cardiovascular lesions to obtain the target features; After the target features are encoded, they are input into the parameter recognition model, and the results of cardiovascular lesion parameter recognition are obtained after processing by the parameter recognition model.

[0005] Optionally, in a first implementation of the first aspect of the present invention, the step of collecting the patient's cardiovascular imaging data, physiological parameter data, and clinical data to obtain multi-source data, and performing noise reduction, missing value and outlier processing on the multi-source data to obtain preprocessed data, includes: Multi-source data is decomposed into sub-band signals of different frequencies by wavelet transform, and then the wavelet-transformed signals are fed into a deep residual shrinkage network to remove high-frequency noise components below a preset threshold. An RNN module is added to the GAIN network. The denoised data is input into the GAIN network, and the RNN module processes the data in chronological order to learn the dependencies and trends between data at adjacent time points. The generator combines the time series information learned by the RNN module to generate missing value data, and the discriminator continuously evaluates the generated data. The data after missing value processing is input into a multimodal adversarial autoencoder, which encodes the data into a low-dimensional feature vector. The decoder reconstructs the data based on the low-dimensional feature vector. If the reconstruction error is abnormal, the current data point is judged to be an outlier and removed, thus obtaining the preprocessed data.

[0006] Optionally, in a second implementation of the first aspect of the present invention, the processing procedure of the deep residual shrinkage network includes: The deep residual shrinkage network contains multiple residual blocks, each of which has an attention mechanism module. The attention mechanism analyzes the signal after wavelet transform to identify the parts related to lesion characteristics and the noise interference area. Based on the analysis results of the attention mechanism, the deep residual shrinkage network determines the threshold through adaptive learning, processes the signal according to the threshold, and removes high-frequency noise components below the threshold to obtain denoised data.

[0007] Optionally, in a third implementation of the first aspect of the present invention, the step of employing a feature selection algorithm based on mutual information to calculate the mutual information value between each feature and cardiovascular lesions to obtain target features includes: The preprocessed data is organized into a feature matrix, where each row represents a patient and each column represents a feature. All features in the feature matrix are included in the initial feature set. For each feature in the initial feature set, the mutual information value between it and cardiovascular lesions is calculated to determine the importance of the feature in predicting cardiovascular lesions. Based on the calculated mutual information values, all features are sorted in descending order. The top N features with the highest mutual information values ​​are selected, and the minimum-maximum normalization method is used to normalize the selected target features to obtain the target features.

[0008] Optionally, in a fourth implementation of the first aspect of the present invention, each row of the feature matrix represents a patient and each column represents a feature.

[0009] Optionally, in a fifth implementation of the first aspect of the present invention, the step of encoding the target features and inputting them into a parameter recognition model, and then processing them through the parameter recognition model to obtain the cardiovascular lesion parameter recognition result, includes: After the target features are encoded, they are input into the parameter recognition model. The multi-head attention mechanism in the spatiotemporal Transformer module is used to process the input features and extract spatiotemporal feature information. The spatiotemporal feature information processed by multi-head spatiotemporal attention is input together with the input features into a gated expert network. The gated expert network processes different types of data features and fuses the processing results according to weights to obtain the fused feature representation. The fused feature representation is input into the Monte Carlo dropout layer, and after multiple repeated forward propagations, the confidence interval of the prediction result is calculated. The results, including predicted values ​​and confidence intervals, are processed and output as cardiovascular lesion parameter identification results.

[0010] Optionally, in a sixth implementation of the first aspect of the present invention, the step of encoding the target features and inputting them into a parameter recognition model, and then processing them through the parameter recognition model to obtain the cardiovascular lesion parameter recognition result, includes: After the target features are encoded, they are input into the parameter recognition model. The multi-head attention mechanism in the spatiotemporal Transformer module is used to process the input features and extract spatiotemporal feature information. The spatiotemporal feature information processed by multi-head spatiotemporal attention is input together with the input features into a gated expert network. The gated expert network processes different types of data features and fuses the processing results according to weights to obtain the fused feature representation. The fused feature representation is input into the Monte Carlo dropout layer, and after multiple repeated forward propagations, the confidence interval of the prediction result is calculated. The results, including predicted values ​​and confidence intervals, are processed and output as cardiovascular lesion parameter identification results.

[0011] A second aspect of the present invention provides a cardiovascular lesion parameter identification system based on data analysis, the system comprising: The data acquisition module is used to collect patients’ cardiovascular imaging data, physiological parameter data and clinical data to obtain multi-source data. The multi-source data is then processed for noise reduction, missing values ​​and outliers to obtain pre-processed data. The feature selection module is used to calculate the mutual information value between each feature and cardiovascular lesions using a mutual information-based feature selection algorithm to obtain the target feature. The parameter recognition module is used to encode the target features and input them into the parameter recognition model. After processing by the parameter recognition model, the cardiovascular lesion parameter recognition results are obtained. The visualization module is used to generate corresponding diagnostic suggestions based on the identification results of cardiovascular lesion parameters, and to visualize the identification results and diagnostic suggestions of cardiovascular lesion parameters.

[0012] A third aspect of the present invention provides a cardiovascular lesion parameter identification device based on data analysis, the device comprising a memory and at least one processor, the memory storing instructions; the at least one processor invokes the instructions in the memory to cause the device to perform the steps of the cardiovascular lesion parameter identification method based on data analysis as described in any of the preceding claims.

[0013] A fourth aspect of the present invention provides a computer-readable storage medium storing instructions that, when executed by a processor, implement the steps of the data analysis-based cardiovascular lesion parameter identification method as described in any of the preceding claims.

[0014] The technical solution provided by this invention involves collecting cardiovascular imaging data, physiological parameter data, and clinical data from patients to obtain multi-source data. Noise reduction, missing value processing, and outlier processing are performed on the multi-source data to obtain preprocessed data. A feature selection algorithm based on mutual information is used to calculate the mutual information value between each feature and the cardiovascular lesion, obtaining the target feature. The target feature is encoded and input into a parameter recognition model, which processes the data to obtain the cardiovascular lesion parameter recognition result. Based on the cardiovascular lesion parameter recognition result, corresponding diagnostic suggestions are generated, and the cardiovascular lesion parameter recognition result and diagnostic suggestions are visualized. This invention employs a multimodal data acquisition method, comprehensively utilizing medical imaging data, physiological parameter data, and clinical information to fully reflect the patient's cardiovascular status, providing a rich data foundation for accurately identifying lesion parameters. This enables accurate and rapid identification of cardiovascular lesion parameters, improving the accuracy and efficiency of cardiovascular disease diagnosis. The output not only intuitively displays the lesion parameter recognition result but also generates corresponding diagnostic suggestions, providing strong support for doctors' clinical decision-making and improving the efficiency and quality of cardiovascular disease diagnosis. Attached Figure Description

[0015] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0016] Figure 1 A flowchart of a method for identifying cardiovascular lesion parameters based on data analysis provided in an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a cardiovascular lesion parameter identification system based on data analysis provided in an embodiment of the present invention; Figure 3 This is a schematic diagram of the structure of a cardiovascular lesion parameter identification device based on data analysis provided in an embodiment of the present invention. Detailed Implementation

[0017] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a particular order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0018] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 The flowchart of the cardiovascular lesion parameter identification method based on data analysis provided in this embodiment of the invention includes the following steps: Step 101: Collect the patient's cardiovascular imaging data, physiological parameter data, and clinical data to obtain multi-source data. Perform noise reduction, missing value and outlier processing on the multi-source data to obtain preprocessed data. In this embodiment, multi-source data is decomposed into sub-band signals of different frequencies using wavelet transform. The wavelet-transformed signals are then fed into a deep residual shrinking network to remove high-frequency noise components below a preset threshold. An RNN module is added to the GAIN network. The denoised data is input into the GAIN network, and the RNN module processes the data in chronological order, learning the dependencies and trends between adjacent time points. The generator combines the time-series information learned by the RNN module to generate missing value data, and the discriminator continuously evaluates the generated data. The missing value processed data is then input into a multimodal adversarial autoencoder, where it is encoded into a low-dimensional feature vector. The decoder reconstructs the data based on the low-dimensional feature vector. If the reconstruction error is abnormal, the current data point is judged as an outlier and removed, resulting in preprocessed data.

[0019] In this embodiment, the deep residual shrinkage network contains multiple residual blocks, each of which is equipped with an attention mechanism module. The attention mechanism analyzes the signal after wavelet transform to determine the parts related to lesion characteristics and noise interference areas. Based on the analysis results of the attention mechanism, the deep residual shrinkage network determines a threshold through adaptive learning. The signal is then processed according to the threshold to remove high-frequency noise components below the threshold, resulting in denoised data.

[0020] Step 102: Using a feature selection algorithm based on mutual information, calculate the mutual information value between each feature and cardiovascular lesions to obtain the target features; In this embodiment, the preprocessed data is organized into a feature matrix, where each row represents a patient and each column represents a feature. All features in the feature matrix are included in an initial feature set. For each feature in the initial feature set, the mutual information value between it and cardiovascular lesions is calculated to determine the importance of the feature in predicting cardiovascular lesions. Based on the calculated mutual information values, all features are sorted in descending order, and the top N features with the highest mutual information values ​​are selected. The min-max normalization method is used to normalize the selected target features to obtain the target features.

[0021] Step 103: After encoding the target features, input them into the parameter recognition model, and obtain the cardiovascular lesion parameter recognition results after processing by the parameter recognition model; In this embodiment, the target features are encoded and input into the parameter recognition model. The input features are processed by the multi-head attention mechanism in the spatiotemporal Transformer module to extract spatiotemporal feature information. The spatiotemporal feature information processed by multi-head spatiotemporal attention is input together with the input features into a gated expert network. The gated expert network processes different types of data features and fuses the processing results according to weights to obtain a fused feature representation. The fused feature representation is input into a Monte Carlo dropout layer, and after multiple repeated forward propagations, the confidence interval of the prediction result is calculated. The results containing the predicted value and the confidence interval are sorted and the cardiovascular lesion parameter recognition result is output.

[0022] In this embodiment, the multi-head attention mechanism analyzes each position in the input features from multiple different perspectives, while the spatiotemporal attention captures the relationship between different regions in the input features in the spatial dimension and mines the changes in the input features over time in the temporal dimension. By calculating the attention weights at different positions, spatiotemporal feature information is extracted.

[0023] In this embodiment, the results of cardiovascular lesion parameter identification are received and converted into medical concept nodes through the UMLS medical knowledge graph. Starting from the mapped medical concept nodes, semantic retrieval is performed in the UMLS medical knowledge graph. The graph attention mechanism is used to calculate the weights of multiple retrieved knowledge paths and nodes, aggregate medical evidence, and form diagnostic suggestions.

[0024] In this embodiment, multimodal data acquisition is performed using medical imaging equipment such as CT scanners, magnetic resonance imaging (MRI) devices, and digital subtraction angiography (DSA) devices to acquire cardiovascular imaging data of patients, including but not limited to two-dimensional and three-dimensional images of the coronary arteries, aorta, and other parts. Simultaneously, physiological parameter data of patients, such as heart rate, blood pressure, electrocardiogram (ECG) data, and blood biochemical indicators (such as blood lipids, blood glucose, and myocardial enzymes), are collected through wearable devices and hospital testing instruments. In addition, clinical information of patients, including medical history, family medical history, and medication history, is also collected.

[0025] In this embodiment, firstly, the acquired electrocardiogram (ECG) signal or medical image data is input into a wavelet transform module. The wavelet transform decomposes the original data into sub-band signals of different frequencies, separating high-frequency noise from low-frequency effective signals. Next, the wavelet-transformed signal is fed into a deep residual contraction network. This network contains multiple residual blocks, each with an attention mechanism module. The attention mechanism analyzes various parts of the signal, determining which regions are important parts related to lesion characteristics and which may be noise interference regions. Based on the analysis results of the attention mechanism, the network adaptively learns... The appropriate threshold is determined by a formula; then the signal is processed according to the threshold to remove high-frequency noise components below the threshold while retaining important signal features related to cardiovascular lesions, thus achieving noise reduction processing of ECG / imaging signals; physiological parameter data containing missing values ​​are input into the generative adversarial interpolation network (GAIN); GAIN consists of two parts: a generator and a discriminator. The generator is responsible for generating possible missing value data, and the discriminator judges the difference between the generated data and the real data; on this basis, in order to adapt to the temporal dependence of physiological parameters, a recurrent neural network (RNN) module is added to the network; The RNN module processes data sequentially, learning the dependencies and trends between adjacent time points. When generating data with missing values, the generator references the temporal information learned by the RNN module to generate data that conforms to the changing patterns of physiological parameters. The discriminator continuously evaluates the generated data and feeds it back to the generator, prompting the generator to optimize the generated data. Through this continuous adversarial training process, the missing values ​​in the data are gradually and accurately filled in. Medical image data and physiological parameter data are simultaneously input into a multimodal adversarial autoencoder (MAAE). MAAE contains multiple encoders and decoders, each corresponding to different modalities. The encoder encodes the medical image data and physiological parameter data into low-dimensional feature vectors, and the decoder then reconstructs the original data based on these feature vectors. During the reconstruction process, the cross-modal reconstruction error is calculated, i.e., comparing the differences before and after reconstruction of the medical image data and the physiological parameter data. If the reconstruction error of a data point in a certain modality is too large, or the difference in reconstruction error between different modalities is abnormal, the data point is judged as an outlier. In this way, MAAE can simultaneously detect grayscale anomalies in medical images and outlier data in physiological parameters.

[0026] In this embodiment, preprocessed physiological parameter data and clinical information are organized into a feature matrix, where each row represents a sample (patient) and each column represents a feature (such as heart rate, blood pressure, specific blood biochemical indicators, etc.). Simultaneously, corresponding label vectors are prepared, with labels indicating whether the patient has cardiovascular disease and the type and severity of the disease. All features in the feature matrix are incorporated into an initial feature set, which contains all features potentially related to cardiovascular disease. For each feature in the initial feature set, its mutual information value with the cardiovascular disease label is calculated. Mutual information is an indicator that measures the dependency between two random variables; by analyzing the correlation between the distribution of feature values ​​and the distribution of disease labels, the effectiveness of the feature in predicting cardiovascular disease is determined. The importance of the lesion is considered; during the calculation process, different types of features (such as continuous and discrete) are considered, and corresponding methods are used for mutual information estimation; based on the calculated mutual information values, all features are sorted in descending order; a threshold is set or the top N features with the highest mutual information values ​​are selected, and these features are filtered out from the initial feature set to form the target feature set; these target features are the features with the strongest correlation with cardiovascular lesions and the most predictive value; the selected target features are normalized by using the min-max normalization method to map the value of each feature to the [0,1] interval; this can eliminate the dimensional differences between different features, so that the subsequent model processing can treat each feature more fairly, and improve the training effect and recognition accuracy of the model.

[0027] In this embodiment, the target features, after being screened and normalized, are encoded and transformed. For numerical physiological and clinical features, they are converted into vector forms suitable for model input according to preset encoding rules, such as mapping continuous numerical features to a vector space of a specific dimension. For image features, the extracted vascular morphology, plaque, and other feature information are encoded into fixed-length feature vectors to ensure that all target features are presented in a unified encoding format, preparing for subsequent input parameter recognition models. The encoded image feature sequence is then input into the spatiotemporal Transformer module. Within this module, a multi-head spatiotemporal attention mechanism processes the image sequence. The image processing involves multi-head attention, which analyzes each position in the image sequence from multiple perspectives, with each "head" focusing on different features and relationships within the sequence. Spatiotemporal attention captures the relationships between different regions in the image spatially and mines the temporal changes in the image sequence. By calculating attention weights between different image frames and at different positions within the same frame, it focuses on key spatiotemporal regions related to cardiovascular lesions, thereby extracting more representative spatiotemporal features. The image features processed by multi-head spatiotemporal attention, along with encoded physiological and clinical features, are then input into the gated expert network of the hybrid expert system. The gated expert network consists of multiple distinct "expert" subnetworks, each adept at handling different types or aspects of data features. The gating mechanism assigns different weights to each "expert" subnetwork based on the characteristics of the input data, determining which subnetworks are more valuable to the current input data. Then, the outputs of different subnetworks are fused according to their weights, achieving effective integration of multi-source data to obtain a fused feature representation. This fused feature representation is then fed into a model trained and optimized with uncertainty awareness. During this process, the integrated Monte Carlo Dropout will, at a certain rate, advance the model's forward propagation. The model randomly "discards" neurons with varying probabilities, and through repeated forward propagation, obtains multiple different outputs to simulate the uncertainty of the model. Evidence-based deep learning then calculates the confidence intervals of the predicted results based on these different outputs. The model comprehensively considers this information, processes the input features, and outputs a result containing the predicted value and the corresponding confidence interval. The results obtained through the above processing, which include the predicted value and the confidence interval, are then organized, and the predicted value of the cardiovascular lesion parameter is output as the recognition result. At the same time, the corresponding confidence interval information is also output to provide doctors with a reference regarding the reliability of the recognition result, thus completing the recognition process of the cardiovascular lesion parameter.

[0028] In this embodiment, the system receives cardiovascular lesion parameter identification results, including numerical and classification information such as the degree of vascular stenosis, plaque characteristics, and lesion type. The medical terms and parameters in these identification results are converted into standardized medical concept nodes in the UMLS medical knowledge graph through a concept mapping mechanism. For example, "50% vascular stenosis" is mapped to the concept of "moderate coronary artery stenosis" in the knowledge graph, establishing a semantic association between the identification results and the knowledge graph. Starting from the mapped medical concept nodes, a semantic search is performed in the UMLS medical knowledge graph. Through relational edges in the knowledge graph (such as "causal relationship," "treatment relationship," and "complication relationship"), clinical guidelines and treatments related to the current lesion parameters are retrieved. The system gathers medical knowledge such as treatment plans, drug information, and prognostic assessments. Employing a graph attention mechanism, it calculates weights for multiple retrieved knowledge paths and nodes, aggregating the most relevant medical evidence to form the knowledge base for diagnostic recommendations. This aggregated medical evidence is then input into a generative model that integrates the knowledge graph. The model performs reasoning and decision-making based on pre-trained language generation capabilities and the structured information of the knowledge graph. First, it matches the most likely diagnostic type based on lesion parameters and clinical guidelines. Then, for that diagnostic type, it generates personalized treatment recommendations, including drug selection, surgical indications, and rehabilitation plans, based on individual patient characteristics (such as age, medical history, and allergy history). During the generation process, the model utilizes the constraints of the knowledge graph. To ensure the medical rationale of the recommendations, key parameters (such as 3D vascular structure, stenosis location, and plaque volume) and diagnostic recommendation text information are extracted from the identification results and preprocessed for visualization. For numerical parameters, normalization and grading are performed; for text recommendations, keywords and structured information are extracted. Simultaneously, relevant anatomical atlases, pathological diagrams, and other visualization resources are obtained from the knowledge graph to prepare for multimodal visualization. 3D medical visualization technology is used to present the spatial structural information of cardiovascular lesions (such as vascular trees and plaque distribution) in an interactive 3D model format, supporting operations such as rotation, scaling, and sectioning. Quantitative indicators of lesion parameters (such as stenosis percentage and hemodynamic parameters) are overlaid and displayed on the same interface. The system uses charts and heatmaps for visualization; it structures and formats the generated diagnostic suggestions, linking them to the visualization model via hyperlinks. When a user clicks on a specific area of ​​the model, relevant diagnostic explanations and treatment suggestions are automatically displayed, forming a richly illustrated diagnostic report. Interactive features are added to the visualization interface, allowing doctors to explore diagnostic results by selecting different parameters and adjusting their viewing angle. When doctors interact with the system, the generative model retrieves relevant background knowledge and evidence from the knowledge graph in real time based on the current focus, generating dynamic explanatory text. For example, when a doctor zooms in on a plaque, the system automatically displays information such as the pathological characteristics, development risks, and treatment priorities of that type of plaque, enhancing the interpretability of the diagnostic suggestions.

[0029] In this embodiment, the cardiovascular lesion parameter identification results cover the location, morphology, degree, functional impact, and auxiliary decision-making information of the lesion; vascular morphological characteristics: degree of vascular stenosis (e.g., percentage of stenosis in the left anterior descending coronary artery), abnormal vessel diameter (specific location of local dilation / contraction, comparison of vessel diameter value with normal reference value), abnormal vessel course (torsion, tortuosity, and assessment of impact on hemodynamics), and vascular wall calcification score (Agatston score or volume score based on CT images); plaque characteristics: plaque location (specific vascular segment, such as the proximal segment of the right coronary artery), plaque volume / burden (quantitative volume value after three-dimensional reconstruction), plaque composition (qualitative / semi-quantitative description of lipid core, fibrous cap thickness, and proportion of calcified components), and plaque vulnerability assessment (plaque rupture risk grading based on imaging characteristics); cardiac function indicators: left ventricular ejection fraction (LVEF, percentage calculated by ultrasound or MRI), diastolic function parameters (e.g., mitral valve flow spectrum E / A ratio, left atrial volume index), and right ventricular function. Functional indicators (right ventricular area change rate, estimated pulmonary artery systolic pressure); hemodynamic status: fractional flow reserve (FFR, numerical value simulated by pressure guidewire or computational fluid dynamics), myocardial blood flow (MBF, quantitative value based on PET or CT perfusion imaging), collateral circulation classification (assessing the degree of patency of compensatory vessels around stenotic vessels); lesion type determination: coronary artery disease (stable / unstable angina, myocardial infarction), aortic disease (aneurysm, dissection, intramural hematoma), cardiomyopathy (hypertrophic, dilated, restrictive), congenital cardiovascular malformations (such as coronary artery origin abnormalities); risk stratification results: acute cardiovascular event risk scores (such as GRACE score, DAPT score), prognostic predictive indicators (5-year cardiovascular mortality risk probability, recurrent myocardial infarction probability); prediction confidence interval: uncertainty range of key parameters (such as vascular stenosis degree: 55%±3%, 95% confidence interval), probability value of lesion type determination (such as "acute myocardial infarction" prediction probability: 92%).

[0030] Please see Figure 2 A schematic diagram of the structure of a cardiovascular lesion parameter identification system based on data analysis provided in this embodiment of the invention. The system includes: The data acquisition module is used to collect patients’ cardiovascular imaging data, physiological parameter data and clinical data to obtain multi-source data. The multi-source data is then processed for noise reduction, missing values ​​and outliers to obtain pre-processed data. The feature selection module is used to calculate the mutual information value between each feature and cardiovascular lesions using a mutual information-based feature selection algorithm to obtain the target feature. The parameter recognition module is used to encode the target features and input them into the parameter recognition model. After processing by the parameter recognition model, the cardiovascular lesion parameter recognition results are obtained.

[0031] Figure 3 This is a schematic diagram of a cardiovascular lesion parameter identification device based on data analysis provided in an embodiment of the present invention. The cardiovascular lesion parameter identification device 600 based on data analysis can vary significantly due to different configurations or performance. It may include one or more central processing units (CPUs) 610 (e.g., one or more processors) and a memory 620, and one or more storage media 630 (e.g., one or more mass storage devices) storing application programs 633 or data 632. The memory 620 and storage media 630 can be temporary or persistent storage. The program stored in the storage media 630 may include one or more modules (not shown in the diagram), each module may include a series of instruction operations on the cardiovascular lesion parameter identification device 600 based on data analysis. Furthermore, the processor 610 may be configured to communicate with the storage media 630 and execute the series of instruction operations in the storage media 630 on the cardiovascular lesion parameter identification device 600 based on data analysis to implement the method provided in the above embodiment.

[0032] The cardiovascular lesion parameter identification device 600 based on data analysis may also include one or more power supplies 640, one or more wired or wireless network interfaces 650, one or more input / output interfaces 660, and / or one or more operating devices 631, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc. Those skilled in the art will understand that... Figure 3 The structure of the cardiovascular lesion parameter identification device based on data analysis shown does not constitute a limitation on the computer device provided by the present invention. It may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0033] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the various steps of the cardiovascular lesion parameter identification method based on data analysis provided in the above embodiments.

[0034] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working process of the above-described equipment or apparatus / unit can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0035] If the integrated unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0036] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for identifying cardiovascular lesion parameters based on data analysis, characterized in that, The method includes the following steps: Collect patients’ cardiovascular imaging data, physiological parameter data and clinical data to obtain multi-source data. Perform noise reduction, missing value and outlier processing on the multi-source data to obtain preprocessed data. A feature selection algorithm based on mutual information is used to calculate the mutual information value between each feature and cardiovascular lesions to obtain the target features; After the target features are encoded, they are input into the parameter recognition model, and the results of cardiovascular lesion parameter recognition are obtained after processing by the parameter recognition model.

2. The method for identifying cardiovascular lesion parameters based on data analysis as described in claim 1, characterized in that, The patient's cardiovascular imaging data, physiological parameter data, and clinical data are collected to obtain multi-source data. Noise reduction, missing value and outlier processing are performed on the multi-source data to obtain preprocessed data, including: Multi-source data is decomposed into sub-band signals of different frequencies by wavelet transform, and then the wavelet-transformed signals are fed into a deep residual shrinkage network to remove high-frequency noise components below a preset threshold. An RNN module is added to the GAIN network. The denoised data is input into the GAIN network, and the RNN module processes the data in chronological order to learn the dependencies and trends between data at adjacent time points. The generator combines the time series information learned by the RNN module to generate missing value data, and the discriminator continuously evaluates the generated data. The data after missing value processing is input into a multimodal adversarial autoencoder, which encodes the data into a low-dimensional feature vector. The decoder reconstructs the data based on the low-dimensional feature vector. If the reconstruction error is abnormal, the current data point is judged to be an outlier and removed, thus obtaining the preprocessed data.

3. The method for identifying cardiovascular lesion parameters based on data analysis as described in claim 2, characterized in that, The processing steps of the deep residual shrinkage network include: The deep residual shrinkage network contains multiple residual blocks, each of which has an attention mechanism module. The attention mechanism analyzes the signal after wavelet transform to identify the parts related to lesion characteristics and the noise interference area. Based on the analysis results of the attention mechanism, the deep residual shrinkage network determines the threshold through adaptive learning, processes the signal according to the threshold, and removes high-frequency noise components below the threshold to obtain denoised data.

4. The method for identifying cardiovascular lesion parameters based on data analysis as described in claim 1, characterized in that, The algorithm employs a mutual information-based feature selection method to calculate the mutual information value between each feature and cardiovascular lesions, thereby obtaining target features, including: The preprocessed data is organized into a feature matrix; All features in the feature matrix are included in the initial feature set. For each feature in the initial feature set, the mutual information value between it and cardiovascular lesions is calculated to determine the importance of the feature in predicting cardiovascular lesions. Based on the calculated mutual information values, all features are sorted in descending order. The top N features with the highest mutual information values ​​are selected, and the minimum-maximum normalization method is used to normalize the selected target features to obtain the target features.

5. The method for identifying cardiovascular lesion parameters based on data analysis as described in claim 4, characterized in that, In the feature matrix, each row represents a patient and each column represents a feature.

6. The method for identifying cardiovascular lesion parameters based on data analysis as described in claim 1, characterized in that, The process of encoding the target features and inputting them into the parameter recognition model, followed by processing by the parameter recognition model to obtain the cardiovascular lesion parameter recognition result, includes: After the target features are encoded, they are input into the parameter recognition model. The multi-head attention mechanism in the spatiotemporal Transformer module is used to process the input features and extract spatiotemporal feature information. The spatiotemporal feature information processed by multi-head spatiotemporal attention is input together with the input features into a gated expert network. The gated expert network processes different types of data features and fuses the processing results according to weights to obtain the fused feature representation. The fused feature representation is input into the Monte Carlo dropout layer, and after multiple repeated forward propagations, the confidence interval of the prediction result is calculated. The results, including predicted values ​​and confidence intervals, are processed and output as cardiovascular lesion parameter identification results.

7. The method for identifying cardiovascular lesion parameters based on data analysis as described in claim 6, characterized in that, The process involves encoding the target features and inputting them into the parameter recognition model. The multi-head attention mechanism in the spatiotemporal Transformer module then processes the input features to extract spatiotemporal feature information, including: Multi-head attention analyzes each position in the input features from multiple different perspectives, while spatiotemporal attention captures the relationship between different regions in the input features in the spatial dimension and mines the changes in the input features over time in the temporal dimension. By calculating the attention weights at different positions, spatiotemporal feature information is extracted.

8. A cardiovascular lesion parameter identification system based on data analysis, characterized in that, The system includes: The data acquisition module is used to collect patients’ cardiovascular imaging data, physiological parameter data and clinical data to obtain multi-source data. The multi-source data is then processed for noise reduction, missing values ​​and outliers to obtain pre-processed data. The feature selection module is used to calculate the mutual information value between each feature and cardiovascular lesions using a mutual information-based feature selection algorithm to obtain the target feature. The parameter recognition module is used to encode the target features and input them into the parameter recognition model. After processing by the parameter recognition model, the cardiovascular lesion parameter recognition results are obtained.

9. A cardiovascular lesion parameter identification device based on data analysis, characterized in that, The data analysis-based cardiovascular lesion parameter identification device includes a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the data analysis-based cardiovascular lesion parameter identification device to perform each step of the data analysis-based cardiovascular lesion parameter identification method as described in any one of claims 1-7.

10. A computer-readable storage medium storing instructions thereon, characterized in that, When the instructions are executed by the processor, they implement the various steps of the data analysis-based cardiovascular lesion parameter identification method as described in any one of claims 1-7.

Images