Construction method and application of ICIs-related myocarditis patient diagnosis model

Abstract

The invention relates to the technical field of medical information processing, and particularly discloses a construction method and application of an ICIs-related myocarditis patient diagnosis model, and the method comprises the steps: collecting multi-modal clinical data of a plurality of ICIs-treated patients, including patient information, diagnosis information, continuous blood biochemical index data and series of heart examination data, the data include electrocardiograms at different time nodes, real-time dynamic cardiac magnetic resonance imaging images and continuous myocardial zymogram index data. Basic and time-varying feature parameters are extracted from blood biochemical and cardiac examination data. And fusing the patient information, the diagnosis information, the basic characteristic parameters and the time-varying characteristic parameters by using a tensor recombination method to obtain effective modeling data. Deep learning and optimization are carried out through a recurrent neural network to obtain an effective diagnosis model; the accuracy and reliability of diagnosis of ICIs-related myocarditis patients are improved, early discovery and treatment are facilitated, and prognosis of the patients is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of medical information processing, and in particular to a method for constructing a diagnostic model for ICIs-related myocarditis patients and its application. Background Art

[0002] At present, immune checkpoint inhibitors (ICIs) are the core means of tumor immunotherapy, and their clinical application is becoming more and more extensive, but the immune-related myocarditis (ICI-myocarditis) caused by them has become the main adverse reaction that threatens the patient's life. At present, clinical diagnosis mainly relies on the threshold judgment of single modality indicators such as myocardial enzyme spectrum (such as troponin), electrocardiogram (ECG) abnormalities and cardiac magnetic resonance imaging (CMR), and refers to the clinical guidelines formulated by the European Society of Cardiology (ESC). In recent years, some studies have attempted to build prediction models by integrating multi-dimensional data, such as combining static clinical parameters with imaging features, or using machine learning algorithms (such as logistic regression and random forests) to analyze the correlation between blood biochemical indicators and cardiac events. In terms of time series data processing, existing technologies mostly use sliding window methods to extract ECG signal features, or use convolutional neural networks (CNN) to analyze CMR images at a single time point. In addition, for the longitudinal changes of myocardial enzyme spectrum, traditional methods mainly rely on linear regression or statistical significance tests. Existing models mostly focus on the static analysis of a single data modality (such as only images or only biochemical indicators), or use simple feature splicing methods to integrate multi-source data, and lack the ability to deeply model the dynamic evolution of time series.

[0003] However, existing methods mostly use biochemical indicators or imaging data at a single time node, and fail to fully integrate the patient's basic information, continuous blood indicators, and the time-series dynamic information of serial cardiac examinations, making it difficult to capture early subtle pathological changes. For example, ECG signals only extract basic parameters such as the RR interval, ignoring the time-varying characteristics of the waveform morphology; CMR image analysis is limited to a single time point, and the evolution of dynamic enhancement patterns is not explored. Traditional feature fusion methods (such as principal component analysis or simple concatenation) cannot effectively handle the spatiotemporal correlation of multi-source heterogeneous data. For example, the nonlinear relationship between static patient age information and dynamic myocardial enzyme spectrum fluctuations has not been modeled, and there is a lack of collaborative analysis mechanism between CMR image features at different time nodes and ECG signals at the same time. Existing deep learning models (such as CNN) have limited processing capabilities for time series data, and it is difficult to synchronously optimize the long-term dependencies of multimodal data. For example, the asynchronous changes in biochemical indicators and imaging manifestations during the development of myocarditis (such as troponin appearing earlier than CMR abnormalities) have not been effectively incorporated into model training, resulting in reduced diagnostic sensitivity. Models based on radiomics rely on high-dimensional feature engineering, require manual labeling of ROI areas, and have a large computational load; lightweight models (such as decision trees) ignore the tensor structure characteristics of multimodal data, and their performance in cross-center data validation is significantly reduced.

[0004] Therefore, the present invention proposes a method for constructing a diagnostic model for patients with ICIs-related myocarditis and its application. Summary of the invention

[0005] The present invention provides a method for constructing a diagnostic model for ICIs-related myocarditis patients and its application, including: collecting multimodal clinical data, covering comprehensive information of patients, and providing a rich data basis for accurate diagnosis. Extracting basic characteristic parameters and time-varying characteristic parameters, continuously collecting blood biochemical indicators and serialized cardiac examination data, and extracting time-varying features (such as wavelet transform coefficients of ECG signals and pixel intensity gradient change rate of CMR images), completely covering the dynamic trajectory of disease development, and being able to capture key features and dynamic changes in the data. High-order tensor reorganization is used to map the basic information of patients (such as age, cancer type) with multimodal time series features in a multidimensional space, retaining nonlinear interactions between data. Data is fused by a tensor reorganization method, achieving effective integration of different types of features and improving data availability. Learning and optimization using a deep learning recurrent neural network model makes the diagnostic model more intelligent and accurate. The accuracy and reliability of diagnosis of patients with ICIs-related myocarditis are improved, which helps early detection and treatment and improves the prognosis of patients.

[0006] The present invention provides a method for constructing a diagnostic model for ICIs-related myocarditis patients, comprising: S1: Collect multimodal clinical data of patients treated with multiple ICIs, where the multimodal clinical data includes patient information, diagnostic information, continuous blood biochemical index data, and serial cardiac examination data. The serial cardiac examination data includes electrocardiogram signal data at different time points, real-time dynamic cardiac magnetic resonance imaging images, and continuous myocardial enzyme spectrum index data; S2: Extract basic characteristic parameters and time-varying characteristic parameters from continuous blood biochemical index data and serial cardiac examination data; S3: Based on the tensor recombination method, the basic characteristic parameters of patient information, diagnostic information and serialized cardiac examination data as well as the time-varying characteristic parameters are integrated to obtain effective modeling data; S4: Use a deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective diagnostic model for ICIs-related myocarditis patients.

[0007] Preferably, the method for constructing a diagnostic model for ICIs-related myocarditis patients, S2: extracting basic characteristic parameters and time-varying characteristic parameters from continuous blood biochemical index data and serialized cardiac examination data, including: Calculating a time-varying characteristic parameter of the blood biochemical index based on the continuous blood biochemical index data as a first time-varying characteristic parameter; Extracting basic characteristic parameters and second time-varying characteristic parameters from serialized cardiac examination data; The time-varying characteristic parameters include a first time-varying characteristic parameter and a second time-varying characteristic parameter.

[0008] Preferably, the method for constructing a diagnostic model for ICIs-related myocarditis patients extracts basic characteristic parameters and second time-varying characteristic parameters from serialized cardiac examination data, including: Based on a method combining wavelet transform and empirical mode decomposition, the electrocardiogram signal data in the serialized heart examination data is decomposed into intrinsic mode functions of multiple frequency components, and characteristic parameters of the intrinsic mode function of each frequency component are extracted as the first basic characteristic parameters; Segmenting a plurality of cardiac tissue dynamic images from the real-time dynamic cardiac magnetic resonance imaging images in the serialized cardiac examination data, extracting the morphological features and magnetic resonance parameter features of the plurality of cardiac tissue dynamic images as the second basic feature parameters, and extracting the morphological time-varying features of the plurality of cardiac tissue dynamic images as the first sub-time-varying feature parameters; Calculate the time-varying characteristic parameter of the myocardial enzyme spectrum index based on the continuous myocardial enzyme spectrum index data in the serialized heart examination data as the second sub-time-varying characteristic parameter; Wherein, the basic characteristic parameters include a first basic characteristic parameter and a second basic characteristic parameter; The second time-varying characteristic parameter includes a first sub-time-varying characteristic parameter and a second sub-time-varying characteristic parameter.

[0009] Preferably, the method for constructing a diagnostic model for ICIs-related myocarditis patients, S3: fusing basic characteristic parameters and time-varying characteristic parameters of patient information, diagnostic information and serialized cardiac examination data based on a tensor recombination method to obtain effective modeling data, including: Arrange the basic characteristic parameters and time-varying characteristics of patient information, diagnostic information and serialized cardiac examination data into multiple characteristic parameter matrices, and combine all characteristic parameter matrices into a three-dimensional tensor; Decompose the three-dimensional tensor into a core tensor and three factor matrices based on a preset objective function; The core tensor and three factor matrices are recombined to obtain effective modeling data.

[0010] Preferably, the method for constructing a diagnostic model for ICIs-related myocarditis patients defines a preset objective function as: In the formula, is a three-dimensional tensor, is the core tensor, , , There are three factor matrices, Indicates The pattern favors tensor-matrix products, represents the Frobenius norm, Indicates mixed norm, and is the regularization parameter.

[0011] Preferably, the method for constructing a diagnostic model for ICIs-related myocarditis patients recombines the core tensor and the three factor matrices to obtain effective modeling data, including: In the formula, is the combined tensor; The combined tensor is expanded into a two-dimensional matrix form to obtain effective modeling data.

[0012] Preferably, the method for constructing a diagnostic model for patients with ICIs-related myocarditis, S4: using a deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective diagnostic model for patients with ICIs-related myocarditis, comprising: A deep learning recurrent neural network model was used to input valid modeling data into the model in time series for training. During the training process, the cross entropy loss function was used to calculate the error between the predicted value and the true value. The adaptive moment estimation optimizer was used to dynamically adjust the model parameters until the initial ICIs-related myocarditis patient diagnosis model was obtained. An independent test data set was used to evaluate the initial diagnostic model, and based on the evaluation results, a random search algorithm was used to adjust the model hyperparameters for training and evaluation again until an effective diagnostic model for ICIs-related myocarditis patients was obtained.

[0013] The present invention provides a system for constructing a diagnostic model for patients with ICIs-related myocarditis, which is used to execute any of the above methods for constructing a diagnostic model for patients with ICIs-related myocarditis, including: A data collection module is used to collect multimodal clinical data of patients treated with multiple ICIs, wherein the multimodal clinical data includes patient information, diagnostic information, continuous blood biochemical index data, and serialized cardiac examination data. The serialized cardiac examination data includes electrocardiogram signal data at different time points, real-time dynamic cardiac magnetic resonance imaging images, and continuous myocardial enzyme spectrum index data; A feature extraction module is used to extract basic feature parameters and time-varying feature parameters from continuous blood biochemical index data and serialized cardiac examination data; A tensor reorganization module is used to fuse basic characteristic parameters and time-varying characteristic parameters of patient information, diagnostic information and serialized cardiac examination data based on a tensor reorganization method to obtain effective modeling data; The deep learning module is used to use the deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective ICIs-related myocarditis patient diagnosis model.

[0014] The present invention provides a device for constructing a diagnostic model for patients with ICIs-related myocarditis, comprising: Memory and processor; The memory is used to store program instructions; The processor is used to call program instructions, and when the program instructions are executed, it is used to perform the following operations: Obtain multimodal data of patients to be diagnosed; The multimodal data of the patient to be diagnosed is input into the effective ICIs-related myocarditis patient diagnostic model constructed based on any of the above methods for constructing an ICIs-related myocarditis patient diagnostic model to obtain the diagnostic information of the patient to be diagnosed.

[0015] The present invention provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the following steps are implemented: Obtain multimodal data of patients to be diagnosed; The multimodal data of the patient to be diagnosed is input into the effective ICIs-related myocarditis patient diagnostic model constructed based on any of the above methods for constructing an ICIs-related myocarditis patient diagnostic model to obtain the diagnostic information of the patient to be diagnosed.

[0016] The beneficial effects of the present invention compared with the prior art are: collecting multimodal clinical data, covering the comprehensive information of patients, and providing a rich data basis for accurate diagnosis. Extracting basic characteristic parameters and time-varying characteristic parameters, continuously collecting blood biochemical indicators and serialized cardiac examination data, and extracting time-varying features (such as wavelet transform coefficients of ECG signals, pixel intensity gradient change rate of CMR images), completely covering the dynamic trajectory of disease development, and being able to capture key features and dynamic changes in the data. High-order tensor reorganization is used to map the basic information of patients (such as age, cancer type) with multimodal time series features in multidimensional space, retaining nonlinear interactions between data. Data is fused by tensor reorganization method, achieving effective integration of different types of features and improving data availability. Learning and optimization using deep learning recurrent neural network model makes the diagnostic model more intelligent and accurate. The accuracy and reliability of diagnosis of patients with ICIs-related myocarditis are improved, which helps early detection and treatment and improves the prognosis of patients.

[0017] Other features and advantages of the present invention will be described in the following description, and partly become apparent from the description, or understood by practicing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures specifically pointed out in this application document.

[0018] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 is a flow chart of a method for constructing a diagnostic model for ICIs-related myocarditis patients in an embodiment of the present invention; Figure 2 Schematic diagram of a system for constructing a diagnostic model for ICIs-related myocarditis patients in an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0021] Embodiment 1: The present invention provides a method for constructing a diagnostic model for ICIs-related myocarditis patients, referring to Figure 1 ,include: S1: Collect multimodal clinical data of patients treated with multiple ICIs, where the multimodal clinical data includes patient information, diagnostic information, continuous blood biochemical index data, and serial cardiac examination data. The serial cardiac examination data includes electrocardiogram signal data at different time points, real-time dynamic cardiac magnetic resonance imaging images, and continuous myocardial enzyme spectrum index data; S2: Extract basic characteristic parameters and time-varying characteristic parameters from continuous blood biochemical index data and serial cardiac examination data; S3: Based on the tensor recombination method, the basic characteristic parameters of patient information, diagnostic information and serialized cardiac examination data as well as the time-varying characteristic parameters are integrated to obtain effective modeling data; S4: Use a deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective diagnostic model for ICIs-related myocarditis patients.

[0022] In this embodiment, ICIs-treated patients refer to patients who are receiving treatment with immune checkpoint inhibitors.

[0023] In this embodiment, the patient information includes the patient's age, gender, underlying diseases and other basic personal information.

[0024] In this embodiment, the diagnosis information is the diagnosis result and related basis about the patient's condition.

[0025] In this embodiment, the continuous blood biochemical index data is the relevant data of biochemical substances in the blood obtained by continuous detection over a period of time, such as the troponin level at different times.

[0026] In this embodiment, the serialized heart examination data is data obtained from a series of heart examinations performed at different times.

[0027] In this embodiment, the electrocardiogram signal data at different time nodes, the real-time dynamic cardiac magnetic resonance imaging images, and the continuous myocardial enzyme spectrum index data refer to the electrocardiogram signals collected at different specific times, the real-time cardiac magnetic resonance imaging images, and the continuously detected myocardial enzyme spectrum index data.

[0028] In this embodiment, the basic characteristic parameters are parameters extracted from the original data and can reflect the basic characteristics of the data.

[0029] In this embodiment, the time-varying characteristic parameter is a parameter that can reflect the characteristics of data changing over time.

[0030] In this embodiment, the effective modeling data is high-quality data that has been processed and integrated and can be used to build a diagnostic model.

[0031] In this embodiment, the deep learning recurrent neural network model is a neural network model based on deep learning and capable of processing time series data.

[0032] In this embodiment, the effective ICIs-related myocarditis patient diagnosis model is a model that has been trained and optimized and can accurately diagnose patients with immune checkpoint inhibitor-related myocarditis.

[0033] The beneficial effects of the above technologies are: collecting multimodal clinical data, covering comprehensive information of patients, and providing a rich data foundation for accurate diagnosis. Extracting basic characteristic parameters and time-varying characteristic parameters, continuously collecting blood biochemical indicators and serialized cardiac examination data, and extracting time-varying features (such as wavelet transform coefficients of ECG signals and pixel intensity gradient change rate of CMR images), completely covering the dynamic trajectory of disease development, and being able to capture key features and dynamic changes in the data. High-order tensor reorganization is used to map the basic information of patients (such as age, cancer type) with multimodal time series features in multidimensional space, retaining nonlinear interactions between data. Data fusion through tensor reorganization method realizes effective integration of different types of features and improves data availability. Using deep learning recurrent neural network model for learning and optimization makes the diagnostic model more intelligent and accurate. It improves the accuracy and reliability of diagnosis of patients with ICIs-related myocarditis, helps early detection and treatment, and improves patient prognosis.

[0034] Embodiment 2: Based on Example 1, a method for constructing a diagnostic model for ICIs-related myocarditis patients, S2: extracting basic characteristic parameters and time-varying characteristic parameters from continuous blood biochemical index data and serialized cardiac examination data, including: Calculating a time-varying characteristic parameter of the blood biochemical index based on the continuous blood biochemical index data as a first time-varying characteristic parameter; Extracting basic characteristic parameters and second time-varying characteristic parameters from serialized cardiac examination data; The time-varying characteristic parameters include a first time-varying characteristic parameter and a second time-varying characteristic parameter.

[0035] In this embodiment, calculating the time-varying characteristic parameters of blood biochemical indicators based on continuous blood biochemical indicator data means analyzing and calculating the blood biochemical indicator data (such as troponin levels, liver and kidney function indicators, etc. at different time points) continuously acquired over a period of time to obtain parameters that can reflect the characteristics of these indicators changing over time. For example, it may include the rate of increase or decrease of the indicator, the periodicity of the change, the amplitude of the fluctuation, etc. These parameters help to more deeply understand the dynamic change of the patient's condition in the time dimension.

[0036] The beneficial effects of the above technical solutions include: calculating time-varying characteristic parameters from continuous blood biochemical index data, enriching the characteristic information of blood biochemistry. Extracting basic characteristic parameters and time-varying characteristic parameters from serialized cardiac examination data, comprehensively covering the key features of cardiac examination data. Clearly distinguishing the time-varying characteristic parameters of blood biochemical indicators and the time-varying characteristic parameters of cardiac examination data, making feature extraction clearer and more targeted. It helps to more accurately reflect the dynamic changes and potential laws of the patient's condition. It improves the quality and effectiveness of feature parameter extraction, laying the foundation for the subsequent construction of an accurate diagnostic model.

[0037] Embodiment 3: Based on Example 2, the method for constructing a diagnostic model for ICIs-related myocarditis patients extracts basic characteristic parameters and second time-varying characteristic parameters from serialized cardiac examination data, including: Based on a method combining wavelet transform and empirical mode decomposition, the electrocardiogram signal data in the serialized heart examination data is decomposed into intrinsic mode functions of multiple frequency components, and characteristic parameters of the intrinsic mode function of each frequency component are extracted as the first basic characteristic parameters; Segmenting a plurality of cardiac tissue dynamic images from the real-time dynamic cardiac magnetic resonance imaging images in the serialized cardiac examination data, extracting the morphological features and magnetic resonance parameter features of the plurality of cardiac tissue dynamic images as the second basic feature parameters, and extracting the morphological time-varying features of the plurality of cardiac tissue dynamic images as the first sub-time-varying feature parameters; Calculate the time-varying characteristic parameter of the myocardial enzyme spectrum index based on the continuous myocardial enzyme spectrum index data in the serialized heart examination data as the second sub-time-varying characteristic parameter; Wherein, the basic characteristic parameters include a first basic characteristic parameter and a second basic characteristic parameter; The second time-varying characteristic parameter includes a first sub-time-varying characteristic parameter and a second sub-time-varying characteristic parameter.

[0038] In this embodiment, the method based on the combination of wavelet transform and empirical mode decomposition is used to decompose the electrocardiogram signal data in the serialized heart examination data into the intrinsic mode functions of multiple frequency components: it refers to the use of the combination of wavelet transform and empirical mode decomposition to process the electrocardiogram signal data and disassemble it into intrinsic mode functions with different frequency components, so as to analyze the characteristics of the signal more deeply. The method based on the combination of wavelet transform and empirical mode decomposition is used to decompose the electrocardiogram signal data into the intrinsic mode functions of multiple frequency components, which can be roughly carried out in the following steps: First, wavelet transform is performed. Select a suitable wavelet basis function and decomposition layer number, decompose the original electrocardiogram signal, and obtain wavelet coefficients at different scales. These wavelet coefficients reflect the information of the signal in different frequency bands. Then, empirical mode decomposition is performed. The signal after preliminary processing by wavelet transform is used as input, and the local maximum and minimum values ​​in the signal are found by iterative screening, and the upper and lower envelopes are fitted, and the mean of the envelope is calculated to obtain an intrinsic mode function (IMF). Repeat this process until the remaining signal is a monotonic function or only the trend term remains. By combining these two technologies, multiple intrinsic mode functions with different frequency components can be obtained. These intrinsic mode functions more clearly show the characteristics of the ECG signal at different frequencies, which is helpful for subsequent more in-depth analysis and diagnosis. In actual operation, it is also necessary to preprocess the data (such as denoising, baseline correction, etc.), select appropriate parameters and algorithms to optimize the decomposition effect, and interpret and apply the decomposition results in combination with professional knowledge and clinical needs.

[0039] In this embodiment, the characteristic parameters of the intrinsic mode function of each frequency component are extracted as the first basic characteristic parameters: from the intrinsic mode function of each frequency component obtained by the above decomposition, the parameters that can describe its characteristics are obtained, and these parameters are used as the first type of basic features for subsequent analysis and modeling. The following common methods can be used to extract characteristic parameters from the intrinsic mode function of each frequency component as the first basic characteristic parameters: 1. Calculate the mean: find the average value of the intrinsic mode function on the time axis to reflect its overall level. 2. Calculate the variance: describe the degree of discreteness of the intrinsic mode function and reflect the fluctuation of the signal. 3. Calculate the energy: square the intrinsic mode function and integrate it on the time axis to reflect the strength of the signal. 4. Calculate the peak value: find the maximum value in the intrinsic mode function to understand the prominent part of the signal. 5. Calculate the frequency center of gravity: reflect the frequency distribution center of the eigenmode function. These parameters obtained by the above calculation can effectively describe the characteristics of each intrinsic mode function as the first type of basic characteristic parameters for subsequent analysis and modeling. The specific parameters to be selected and how to calculate them will be adjusted and optimized based on specific research needs and data characteristics.

[0040] In this embodiment, a plurality of dynamic images of cardiac tissue are segmented from the real-time dynamic cardiac magnetic resonance imaging images in the serialized cardiac examination data: the real-time dynamic cardiac magnetic resonance imaging images are processed to separate different cardiac tissue parts to form independent dynamic images.

[0041] In this embodiment, morphological features and magnetic resonance parameter features of various dynamic images of cardiac tissues are extracted as the second basic feature parameters: from the various dynamic images of cardiac tissues obtained by segmentation, morphological features such as shape and size, as well as parameter features related to magnetic resonance, are obtained and used as the second type of basic features. The following methods can be used to extract morphological features and magnetic resonance parameter features from various dynamic images of cardiac tissues as the second basic feature parameters: For morphological features: 1. Shape description: calculate the area, perimeter, circularity (used to describe the degree of closeness to a circle), rectangularity (ratio to the circumscribed rectangle), etc. of cardiac tissue. 2. Size measurement: determine the linear dimensions such as length, width, thickness, and three-dimensional dimensions such as volume of cardiac tissue. For magnetic resonance parameter features: 1. Signal intensity: obtain the signal intensity values ​​of cardiac tissue in different areas of the magnetic resonance image. 2. Relaxation time: including longitudinal relaxation time (T1) and transverse relaxation time (T2), etc. 3. Contrast: For example, the contrast between tissues is used to distinguish different types of cardiac tissues. These features can be extracted and quantified by image processing and analysis algorithms, combined with the physical principles of magnetic resonance imaging and related calculation methods.

[0042] In this embodiment, the morphological time-varying features of multiple dynamic images of cardiac tissue are extracted as the first sub-time-varying feature parameters: the characteristics of the morphological changes of these dynamic images of cardiac tissue over time are obtained and used as the first type of time-varying features. To extract the morphological time-varying features from multiple dynamic images of cardiac tissue as the first sub-time-varying feature parameters, the following steps can be used: 1. Image registration: First, the dynamic images of cardiac tissue at different time points are registered to ensure that they are consistent in spatial position so as to accurately compare the changes in morphology. 2. Morphological measurement: At each time point, the morphology of cardiac tissue is quantitatively measured, such as calculating morphological parameters such as area, circumference, and volume. 3. Change calculation: Compare the differences in morphological parameters at adjacent time points or within a specific time interval, and calculate the amount of change or the rate of change. 4. Feature extraction: These amounts of change or rates of change are extracted as morphological time-varying feature parameters. For example, if the expansion and contraction of the heart ventricle are concerned, the rate of change of the ventricular volume over time can be calculated. 5. Feature selection: According to the research purpose and data characteristics, the most representative and discriminative time-varying features are selected as the final first sub-time-varying feature parameters.

[0043] In this embodiment, the time-varying characteristic parameter of the myocardial enzyme spectrum index is calculated based on the continuous myocardial enzyme spectrum index data in the serialized cardiac examination data as the second sub-time-varying characteristic parameter: according to the myocardial enzyme spectrum index data obtained by continuous detection, the characteristic parameter that can reflect its change over time is calculated as the second type of time-varying feature. To calculate the time-varying characteristic parameter as the second sub-time-varying characteristic parameter based on the continuous myocardial enzyme spectrum index data, it can be achieved in the following ways: 1. Data collection: First, a series of myocardial enzyme spectrum index data detected at different time points are obtained, such as creatine kinase (CK), creatine kinase isoenzyme (CK-MB), lactate dehydrogenase (LDH), etc. 2. Trend analysis: Through linear regression or polynomial fitting and other methods, the overall change trend of these indicators over time is analyzed to obtain the slope of the trend line to reflect the speed of the indicator rising or falling. 3. Fluctuation evaluation: Calculate the standard deviation or variance of the indicator data to measure its degree of fluctuation in the time series. 4. Change amplitude calculation: Determine the change amplitude of the indicator at adjacent time points or in a specific time period, such as calculating the difference or proportional change of the indicator between two time points. 5. Periodicity detection: Use Fourier transform or other spectrum analysis methods to detect whether the index data has a periodic change pattern. These calculated characteristic parameters can reflect the changes of myocardial enzyme spectrum indicators over time and are used as the second type of time-varying characteristic parameters for subsequent analysis and modeling.

[0044] The beneficial effects of the above technical solutions include: using a method combining wavelet transform and empirical mode decomposition to process electrocardiogram signal data, and the extracted intrinsic mode function characteristic parameters more accurately reflect the characteristics of cardiac electrical activity. The characteristics of various cardiac tissue dynamic images are segmented and extracted from real-time dynamic cardiac magnetic resonance imaging images, including morphological and magnetic resonance parameter characteristics and morphological time-varying characteristics, which comprehensively describe the changes in cardiac structure and function. The time-varying characteristic parameters are calculated based on continuous myocardial enzyme spectrum index data, which provides an important basis for evaluating the dynamic changes of myocardial damage. The combination of multiple basic characteristic parameters and time-varying characteristic parameters improves the depth and breadth of analysis of cardiac examination data. The extracted features are made more comprehensive and refined, providing strong support for building a high-precision myocarditis diagnosis model.

[0045] Embodiment 4: Based on Example 1, a method for constructing a diagnostic model for ICIs-related myocarditis patients, S3: Based on the tensor recombination method, basic feature parameters and time-varying feature parameters of patient information, diagnostic information and serialized cardiac examination data are integrated to obtain effective modeling data, including: Arrange the basic characteristic parameters and time-varying characteristics of patient information, diagnostic information and serialized cardiac examination data into multiple characteristic parameter matrices, and combine all characteristic parameter matrices into a three-dimensional tensor; Decompose the three-dimensional tensor into a core tensor and three factor matrices based on a preset objective function; The core tensor and three factor matrices are recombined to obtain effective modeling data.

[0046] In this embodiment, the basic characteristic parameters of the patient information, diagnosis information and serialized cardiac examination data and the time-varying characteristics are organized into a plurality of characteristic parameter matrices: Assume the first basic characteristic parameter matrix is ( Represents the first basic characteristic parameter OK column matrix), where represents the number of samples, represents the dimension of the first basic feature parameter; the second basic feature parameter matrix is ( Represents the second basic characteristic parameter OK Column matrix), similarly is the sample size, is the dimension of the second basic characteristic parameter; the first sub-time-varying characteristic parameter matrix is ( Represents the time-varying characteristic parameter of the first sub OK Column matrix), is the sample size, is the dimension of the first sub-time-varying characteristic parameter; the second sub-time-varying characteristic parameter matrix is ( It represents the time-varying characteristic parameter of the second sub OK Column matrix), is the sample size, is the dimension of the second sub-time-varying characteristic parameter.

[0047] In this embodiment, all feature parameter matrices are combined into a three-dimensional tensor ( is composed of all characteristic parameter matrices OK List rank), where represents the number of samples after unification, represents the total dimension of the basic feature parameters, Represents the total dimension of the time-varying feature parameters.

[0048] The beneficial effects of the above technical solutions include: organizing multiple feature parameters into three-dimensional tensors, achieving unified representation of different types of features, and facilitating subsequent processing. By decomposing the three-dimensional tensor through a preset objective function, the potential structure and key information in the data can be extracted. Recombining the core tensor and factor matrix to obtain effective modeling data improves the integrity and effectiveness of the data. This fusion method reduces the redundancy and noise of the data and highlights important features. It provides a high-quality, structured data foundation for building an accurate diagnostic model.

[0049] Embodiment 5: Based on Example 4, the method for constructing a diagnostic model for ICIs-related myocarditis patients defines the preset objective function as: In the formula, is a three-dimensional tensor, is the core tensor, , , There are three factor matrices, Indicates The pattern favors tensor-matrix products, represents the Frobenius norm, Indicates mixed norm, and is the regularization parameter.

[0050] In this embodiment, in the above fusion process related formula, and It is usually a weight parameter used to balance different items or constraints. Their values ​​need to be determined according to the specific data characteristics, problem requirements and model performance. The following are some examples: Assuming that in a relatively simple scenario, it is hoped that the basic feature parameters and time-varying feature parameters have similar importance in the fusion process, and the fluctuation range of the data is relatively small, it may be taken =0.5, =0.5. In this way, when fusion is performed, the contribution of the basic feature parameters and the time-varying feature parameters to the final result is basically the same. If the basic feature parameters are more important to the model than the time-varying feature parameters, for example, the basic feature parameters can provide more critical diagnostic information, while the time-varying feature parameters only play a certain auxiliary role, it may be necessary to take =0.7, =0.3. In this case, the basic feature parameters will have a greater impact on the fusion process, highlighting the dominant position of basic features in diagnosis. On the contrary, if the time-varying feature parameters are more important in diagnosis, such as the development and changes of the disease are mainly reflected by the time-varying features, then =0.3, =0.7, so that time-varying features occupy a larger proportion in the fusion results. In practical applications, cross-validation, grid search and other methods can also be used to determine and For example, set a range of values, such as , , with a step size of 0.1, and then traverse all possible combinations, calculate the performance indicators of the model under each combination (such as accuracy, recall rate, F1 value, etc.), and select the one that makes the model performance optimal and The value of is taken as the final value.

[0051] In this embodiment, decomposing the three-dimensional tensor into a core tensor and three factor matrices based on the preset objective function includes: fixed , , ,renew : The objective function is Taking the partial derivative and setting it to zero, we get: in Representing a tensor In the first mode of unfolding the matrix, is the identity matrix, Represents the transpose operation of a matrix.

[0052] fixed , , ,renew : Similarly, the objective function is Taking the partial derivative and setting it to zero, we get: in Representing a tensor The expanded matrix in the second mode, Represents the transpose operation of a matrix.

[0053] fixed , , ,renew : The objective function is Taking the partial derivative and setting it to zero, we get: in Representing a tensor The expanded matrix in the third mode, Represents the transpose operation of a matrix.

[0054] fixed , , ,renew : The objective function is Taking the partial derivative and setting it to zero, we get: in It's about the core tensor The regularization term is used to constrain the sparsity of the core tensor. Represents the transpose operation of a matrix.

[0055] The factor matrix and core tensor are updated iteratively until the objective function converges.

[0056] In the above-mentioned fusion process, the first mode, the second mode, and the third mode are a way of defining the meanings represented by different dimensions of tensors. The following is a detailed explanation: The first mode: It can usually be understood as representing different samples, that is, different individual patients. In this mode, each slice of the tensor corresponds to all relevant feature data of a patient. It organizes data from the individual dimension to distinguish the differences between different patients so that the model can learn the relationship between different patient characteristics and the diagnosis of ICIs-related myocarditis.

[0057] The second mode: It can be regarded as representing different types of features. For example, in the serialized cardiac examination data, different types of features are included, such as electrocardiogram signal data features, real-time dynamic cardiac magnetic resonance imaging image features, and continuous myocardial enzyme spectrum index data features. The second mode is to organize data from the dimension of feature type, which helps the model capture the importance and relationship of different feature types for diagnosis.

[0058] The third mode is generally used to represent the time dimension. Because the data is serialized and contains information at different time nodes, the third mode is used to reflect the changes in data over time, such as changes in electrocardiograms and myocardial enzyme spectrum indicators at different time points, so that the model can learn the impact of time-varying features on the diagnosis of ICIs-related myocarditis.

[0059] The beneficial effects of the above technical solutions include: a well-defined preset objective function provides a clear mathematical model and criteria for tensor decomposition and data fusion. By using the Frobenius norm and the mixed norm, the difference between the tensor and the decomposition result and the complexity of the factor matrix can be effectively measured. The introduction of regularization parameters helps to control the complexity of the model, prevent overfitting, and improve the generalization ability of the model. This precise mathematical definition makes the data fusion process more scientific, rigorous and explainable. It provides strong mathematical guarantee and theoretical support for obtaining accurate, reliable and effective modeling data.

[0060] Embodiment 6: Based on Example 1, the method for constructing a diagnostic model for ICIs-related myocarditis patients recombines the core tensor and three factor matrices to obtain effective modeling data, including: In the formula, is the combined tensor; The combined tensor is expanded into a two-dimensional matrix form to obtain effective modeling data.

[0061] In this embodiment, expanding the combined tensor into a two-dimensional matrix form to obtain effective modeling data means expanding the combined tensor obtained by specific calculation or reorganization according to certain rules and algorithms to present it as a two-dimensional matrix structure. The specific implementation may be to determine the order and rules of expansion according to the dimension and element arrangement of the tensor, and map the multi-dimensional tensor elements to the corresponding positions of the two-dimensional matrix, thereby forming an effective data form suitable for subsequent modeling analysis.

[0062] The beneficial effects of the above technical solutions include: the core tensor and factor matrix are recombined through a specific formula to ensure the accuracy and rationality of data fusion. The combined tensor is expanded into a two-dimensional matrix form to facilitate subsequent application and processing in the diagnostic model. This combination method can effectively integrate key information in multimodal data and improve the quality of modeling data. It provides a reliable data foundation for building a high-performance diagnostic model for myocarditis patients. The fusion and representation of data are optimized, which helps to improve the accuracy and efficiency of the diagnostic model.

[0063] Embodiment 7: Based on Example 1, a method for constructing a diagnostic model for patients with ICIs-related myocarditis, S4: using a deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective diagnostic model for patients with ICIs-related myocarditis, including: A deep learning recurrent neural network model was used to input valid modeling data into the model in time series for training. During the training process, the cross entropy loss function was used to calculate the error between the predicted value and the true value. The adaptive moment estimation optimizer was used to dynamically adjust the model parameters until the initial ICIs-related myocarditis patient diagnosis model was obtained. An independent test data set was used to evaluate the initial diagnostic model, and based on the evaluation results, a random search algorithm was used to adjust the model hyperparameters for training and evaluation again until an effective diagnostic model for ICIs-related myocarditis patients was obtained.

[0064] In this embodiment, the cross entropy loss function is used to calculate the error between the predicted value and the true value during the training process, and the model parameters are dynamically adjusted using an adaptive moment estimation optimizer until an initial ICIs-related myocarditis patient diagnosis model is obtained: During the model training phase, the cross entropy loss function is used to measure the degree of difference between the value predicted by the model and the actual true value. At the same time, an adaptive moment estimation optimizer is used to automatically and dynamically adjust various parameters (such as weights, biases, etc.) in the model according to this error, and the model is continuously optimized. After multiple iterations, the model reaches a preliminary usable state, and an initial ICIs-related myocarditis patient diagnosis model is obtained.

[0065] In this embodiment, an independent test data set is used to evaluate the initial diagnostic model, and based on the evaluation results, a random search algorithm is used to adjust the model hyperparameters for training and evaluation again, until an effective ICIs-related myocarditis patient diagnosis model is obtained: a set of test data that is completely independent of the training data is used to test the performance of the initial diagnostic model, and a random search algorithm is used to find and adjust the model's hyperparameters (such as learning rate, number of layers, number of nodes, etc.) based on the indicators obtained by the evaluation (such as accuracy, recall rate, etc.). Then, the new parameters are used for training and evaluation again, and this process is repeated until the performance of the model reaches expectations, thereby obtaining a final effective ICIs-related myocarditis patient diagnosis model.

[0066] The beneficial effects of the above technical solutions include: using a recurrent neural network model to process effective modeling data in time series can fully capture the time dependency in the data. The application of the cross entropy loss function accurately measures the error between the predicted value and the true value, which is helpful for model optimization. The adaptive moment estimation optimizer dynamically adjusts the model parameters to improve the training efficiency and convergence speed. The generalization ability and accuracy of the diagnostic model are ensured by evaluating and adjusting the hyperparameters through independent test data sets. The use of random search algorithms can find the optimal configuration in a larger hyperparameter space. The performance and reliability of the diagnostic model are improved, providing a strong guarantee for the accurate diagnosis of patients with ICIs-related myocarditis.

[0067] Embodiment 8: The present invention provides a system for constructing a diagnostic model for patients with ICIs-related myocarditis, which is used to execute any of the above methods for constructing a diagnostic model for patients with ICIs-related myocarditis, with reference to Figure 2 ,include: A data collection module is used to collect multimodal clinical data of patients treated with multiple ICIs, wherein the multimodal clinical data includes patient information, diagnostic information, continuous blood biochemical index data, and serialized cardiac examination data. The serialized cardiac examination data includes electrocardiogram signal data at different time points, real-time dynamic cardiac magnetic resonance imaging images, and continuous myocardial enzyme spectrum index data; A feature extraction module is used to extract basic feature parameters and time-varying feature parameters from continuous blood biochemical index data and serialized cardiac examination data; A tensor reorganization module is used to fuse basic characteristic parameters and time-varying characteristic parameters of patient information, diagnostic information and serialized cardiac examination data based on a tensor reorganization method to obtain effective modeling data; The deep learning module is used to use the deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective ICIs-related myocarditis patient diagnosis model.

[0068] The beneficial effects of the above technologies are: collecting multimodal clinical data, covering comprehensive information of patients, and providing a rich data foundation for accurate diagnosis. Extracting basic characteristic parameters and time-varying characteristic parameters, continuously collecting blood biochemical indicators and serialized cardiac examination data, and extracting time-varying features (such as wavelet transform coefficients of ECG signals and pixel intensity gradient change rate of CMR images), completely covering the dynamic trajectory of disease development, and being able to capture key features and dynamic changes in the data. High-order tensor reorganization is used to map the basic information of patients (such as age, cancer type) with multimodal time series features in multidimensional space, retaining nonlinear interactions between data. Data fusion through tensor reorganization method realizes effective integration of different types of features and improves data availability. Using deep learning recurrent neural network model for learning and optimization makes the diagnostic model more intelligent and accurate. It improves the accuracy and reliability of diagnosis of patients with ICIs-related myocarditis, helps early detection and treatment, and improves patient prognosis.

[0069] Embodiment 9: The present invention provides a device for constructing a diagnostic model for patients with ICIs-related myocarditis, comprising: Memory and processor; The memory is used to store program instructions; The processor is used to call program instructions, and when the program instructions are executed, it is used to perform the following operations: Obtain multimodal data of patients to be diagnosed; The multimodal data of the patient to be diagnosed is input into the effective ICIs-related myocarditis patient diagnostic model constructed based on the method for constructing an ICIs-related myocarditis patient diagnostic model in any one of Examples 1 to 7 to obtain the diagnostic information of the patient to be diagnosed.

[0070] In this embodiment, the diagnostic information of the patient to be diagnosed refers to the relevant information about whether the patient suffers from ICIs-related myocarditis and the specific conditions of the disease (such as the severity of the disease, possible development trends, etc.) obtained by inputting the multimodal data of the patient who has not yet reached a clear diagnosis conclusion into the established diagnostic model.

[0071] The beneficial effects of the above technical solutions include: by storing program instructions in a memory, the stability and reliability of data and operations are ensured. The processor can call program instructions for operation, realizing the automation and intelligence of the diagnostic process. Acquiring multimodal data of patients to be diagnosed provides a comprehensive source of information for diagnosis. Inputting multimodal data into the constructed diagnostic model can quickly and accurately obtain diagnostic information. This device improves the efficiency and accuracy of diagnosis and reduces the errors and complexity of manual operations. It provides a convenient, efficient and accurate tool for the diagnosis of patients with ICIs-related myocarditis, which helps to detect and treat the disease in a timely manner.

[0072] Embodiment 10: The present invention provides a readable storage medium, on which a program or instruction is stored, and when the program or instruction is executed by a processor, the following steps are implemented: Obtain multimodal data of patients to be diagnosed; The multimodal data of the patient to be diagnosed is input into the effective ICIs-related myocarditis patient diagnostic model constructed based on the method for constructing an ICIs-related myocarditis patient diagnostic model in any one of Examples 1 to 7 to obtain the diagnostic information of the patient to be diagnosed.

[0073] The beneficial effects of the above technical solution include: providing a convenient data storage and processing method so that the relevant programs or instructions can be stably saved. When executed by the processor, it can automatically obtain multimodal data of the patient to be diagnosed to ensure the efficiency of data collection. The multimodal data is input into a specifically constructed diagnostic model to achieve fast and accurate diagnosis, saving diagnostic time. This storage medium helps to promote and apply advanced diagnostic technologies and improve the accessibility of medical services. It provides an efficient and standardized solution for the diagnosis of ICIs-related myocarditis, which is conducive to improving the diagnostic level and medical quality.

[0074] Obviously, those skilled in the art can make various changes and modifications to the present invention without departing from the spirit and scope of the present invention. Thus, if these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A method for constructing a diagnostic model for ICIs-related myocarditis patients, characterized in that: include: S1: Collect multimodal clinical data of patients treated with multiple ICIs, where the multimodal clinical data includes patient information, diagnostic information, continuous blood biochemical index data, and serial cardiac examination data. The serial cardiac examination data includes electrocardiogram signal data at different time points, real-time dynamic cardiac magnetic resonance imaging images, and continuous myocardial enzyme spectrum index data; S2: Extract basic characteristic parameters and time-varying characteristic parameters from continuous blood biochemical index data and serial cardiac examination data; S3: Based on the tensor recombination method, the basic characteristic parameters of patient information, diagnostic information and serialized cardiac examination data as well as the time-varying characteristic parameters are integrated to obtain effective modeling data; S4: Use a deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective diagnostic model for ICIs-related myocarditis patients.

2. The method for constructing a diagnostic model for ICIs-related myocarditis patients according to claim 1, characterized in that: S2: Extract basic characteristic parameters and time-varying characteristic parameters from continuous blood biochemical index data and serialized cardiac examination data, including: Calculating a time-varying characteristic parameter of the blood biochemical index based on the continuous blood biochemical index data as a first time-varying characteristic parameter; Extracting basic characteristic parameters and second time-varying characteristic parameters from serialized cardiac examination data; The time-varying characteristic parameters include a first time-varying characteristic parameter and a second time-varying characteristic parameter.

3. The method for constructing a diagnostic model for ICIs-related myocarditis patients according to claim 2, characterized in that: Extracting basic characteristic parameters and second time-varying characteristic parameters from the serialized cardiac examination data includes: Based on a method combining wavelet transform and empirical mode decomposition, the electrocardiogram signal data in the serialized heart examination data is decomposed into intrinsic mode functions of multiple frequency components, and characteristic parameters of the intrinsic mode function of each frequency component are extracted as the first basic characteristic parameters; Segmenting a plurality of cardiac tissue dynamic images from the real-time dynamic cardiac magnetic resonance imaging images in the serialized cardiac examination data, extracting the morphological features and magnetic resonance parameter features of the plurality of cardiac tissue dynamic images as the second basic feature parameters, and extracting the morphological time-varying features of the plurality of cardiac tissue dynamic images as the first sub-time-varying feature parameters; Calculate the time-varying characteristic parameter of the myocardial enzyme spectrum index based on the continuous myocardial enzyme spectrum index data in the serialized heart examination data as the second sub-time-varying characteristic parameter; Wherein, the basic characteristic parameters include a first basic characteristic parameter and a second basic characteristic parameter; The second time-varying characteristic parameter includes a first sub-time-varying characteristic parameter and a second sub-time-varying characteristic parameter.

4. The method for constructing a diagnostic model for ICIs-related myocarditis patients according to claim 1, characterized in that: S3: Based on the tensor reorganization method, the basic characteristic parameters of patient information, diagnostic information and serialized cardiac examination data as well as the time-varying characteristic parameters are integrated to obtain effective modeling data, including: Arrange the basic characteristic parameters and time-varying characteristics of patient information, diagnostic information and serialized cardiac examination data into multiple characteristic parameter matrices, and combine all characteristic parameter matrices into a three-dimensional tensor; Decompose the three-dimensional tensor into a core tensor and three factor matrices based on a preset objective function; The core tensor and three factor matrices are recombined to obtain effective modeling data.

5. The method for constructing a diagnostic model for ICIs-related myocarditis patients according to claim 4, characterized in that: The preset objective function is defined as: In the formula, is a three-dimensional tensor, is the core tensor, , , There are three factor matrices, Indicates The pattern favors tensor-matrix products, represents the Frobenius norm, Indicates mixed norm, and is the regularization parameter.

6. The method for constructing a diagnostic model for ICIs-related myocarditis patients according to claim 1, characterized in that: The core tensor and three factor matrices are recombined to obtain effective modeling data, including: In the formula, is the combined tensor; The combined tensor is expanded into a two-dimensional matrix form to obtain effective modeling data.

7. The method for constructing a diagnostic model for ICIs-related myocarditis patients according to claim 1, characterized in that: S4: Use a deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective ICIs-related myocarditis patient diagnosis model, including: A deep learning recurrent neural network model was used to input valid modeling data into the model in time series for training. During the training process, the cross entropy loss function was used to calculate the error between the predicted value and the true value. The adaptive moment estimation optimizer was used to dynamically adjust the model parameters until the initial ICIs-related myocarditis patient diagnosis model was obtained. An independent test data set was used to evaluate the initial diagnostic model, and based on the evaluation results, a random search algorithm was used to adjust the model hyperparameters for training and evaluation again until an effective diagnostic model for ICIs-related myocarditis patients was obtained.

8. A system for constructing a diagnostic model for patients with ICIs-related myocarditis, characterized in that: A method for constructing a diagnostic model for ICIs-related myocarditis patients according to any one of claims 1 to 7, comprising: A data collection module is used to collect multimodal clinical data of patients treated with multiple ICIs, wherein the multimodal clinical data includes patient information, diagnostic information, continuous blood biochemical index data, and serialized cardiac examination data. The serialized cardiac examination data includes electrocardiogram signal data at different time points, real-time dynamic cardiac magnetic resonance imaging images, and continuous myocardial enzyme spectrum index data; A feature extraction module is used to extract basic feature parameters and time-varying feature parameters from continuous blood biochemical index data and serialized cardiac examination data; A tensor reorganization module is used to fuse basic characteristic parameters and time-varying characteristic parameters of patient information, diagnostic information and serialized cardiac examination data based on a tensor reorganization method to obtain effective modeling data; The deep learning module is used to use the deep learning recurrent neural network model to perform deep learning and optimization on effective modeling data to obtain an effective ICIs-related myocarditis patient diagnosis model.

9. A device for constructing a diagnostic model for patients with ICIs-related myocarditis, characterized in that: include: Memory and processor; The memory is used to store program instructions; The processor is used to call program instructions, and when the program instructions are executed, it is used to perform the following operations: Obtain multimodal data of patients to be diagnosed; The multimodal data of the patient to be diagnosed is input into the effective ICIs-related myocarditis patient diagnostic model constructed based on the method for constructing an ICIs-related myocarditis patient diagnostic model according to any one of claims 1 to 7 to obtain the diagnostic information of the patient to be diagnosed.

10. A readable storage medium, characterized in that: The readable storage medium stores a program or instruction, and when the program or instruction is executed by the processor, the following steps are implemented: Obtain multimodal data of patients to be diagnosed; The multimodal data of the patient to be diagnosed is input into the effective ICIs-related myocarditis patient diagnostic model constructed based on the method for constructing an ICIs-related myocarditis patient diagnostic model according to any one of claims 1 to 7 to obtain the diagnostic information of the patient to be diagnosed.

Images