Processing method and device of hypertrophic cardiomyopathy risk prediction model

Abstract

Embodiments of the present application relate to a processing method and device of a hypertrophic cardiomyopathy risk prediction model, the method comprising: constructing a hypertrophic cardiomyopathy risk prediction model; constructing a generalized population model dataset as a first dataset, and constructing a specified population model dataset as a second dataset based on specified population characteristics; pre-training the model based on the first dataset first, and then fine-tuning the model based on the second dataset; after the fine-tuning is completed, inputting a myocardial strain parameter set of a measured person meeting the specified population characteristics into the model for prediction and feeding back the prediction result to a user. Through the present application, the prediction accuracy and stability can be improved.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a processing method and apparatus for a risk prediction model of hypertrophic cardiomyopathy. Background Technology

[0002] Hypertrophic cardiomyopathy (HCM) is a non-ischemic cardiomyopathy characterized primarily by myocardial hypertrophy. HCM carries a risk of sudden death and is one of the causes of exercise-induced sudden death. Some HCM patients are asymptomatic in the early stages, or even experience sudden death as their first symptom; therefore, early diagnosis of HCM is crucial.

[0003] Cardiac magnetic resonance (CMR) imaging-based myocardial strain measurement is a sensitive and non-invasive method for measuring myocardial strain parameters, enabling quantitative assessment of a range of myocardial strain parameters in the left ventricle (left atrium and left ventricle). Based on the myocardial strain assessment results, early myocardial dysfunction can be identified before a decrease in left ventricular ejection fraction (LVEF). In other words, early risk prediction for hypertrophic cardiomyopathy can be made based on myocardial strain parameters.

[0004] Analysis of traditional hypertrophic cardiomyopathy (HCM) risk prediction schemes revealed several issues: 1) Most schemes only consider left ventricular parameters, neglecting left atrial parameters. Furthermore, left ventricular parameters are primarily based on three categories: global longitudinal strain (GLS), global circumferential strain (GCS), and global radial strain (GRS), without considering other subcategories. Under these circumstances, the accuracy of HCM risk prediction is limited by the parameter dimensions, making further improvement difficult. 2) Although myocardial strain technology can provide precise myocardial strain parameters, the interpretation of its results requires physicians with extensive professional knowledge and experience. In other words, the stability of HCM risk prediction is limited by individual professional experience, making it difficult to obtain stable prediction quality.

[0005] To address these issues, this invention presents a deep learning model based on artificial intelligence technology that can predict the risk of hypertrophic cardiomyopathy based on multiple myocardial strain parameters of the left atrium and left ventricle. This model improves prediction accuracy by leveraging multi-dimensional parameters and overcomes the limitations of personal experience, thereby enhancing prediction stability. The multiple myocardial strain parameters mentioned here include six additional parameters besides the traditional three left ventricular parameters (GLS, GCS, GRS): circumferential transmural strain difference (cTSD), peak systolic strain rate (PSSR), peak diastolic strain rate (PDSR), passive strain (Ee), early diastolic negative strain rate (sRe), and late diastolic negative strain rate (sRa). Summary of the Invention

[0006] The purpose of this invention is to address the shortcomings of existing technologies by providing a method, apparatus, electronic device, and computer-readable storage medium for processing a hypertrophic cardiomyopathy (HCM) risk prediction model. This invention constructs a HCM risk prediction model capable of predicting risk based on six types of left ventricular parameters (GLS, GCS, GRS, cTSD, PSSR, and PDSR) and three types of left atrial parameters (Ee, sRe, and sRa). It also constructs a generalized population model dataset (first dataset) and a specified population model dataset (second dataset) through big data collection. The HCM risk prediction model is first pre-trained based on the first dataset and then fine-tuned based on the second dataset. After fine-tuning, the HCM risk prediction model is used to predict the risk based on the myocardial strain parameters of the subjects. This invention improves prediction accuracy and stability.

[0007] To achieve the above objectives, a first aspect of the present invention provides a method for processing a risk prediction model for hypertrophic cardiomyopathy, the method comprising:

[0008] A deep learning model is constructed to predict the risk probability of hypertrophic cardiomyopathy, denoted as the corresponding hypertrophic cardiomyopathy risk prediction model. This model predicts the risk probability of hypertrophic cardiomyopathy based on the input myocardial strain parameter vector X and outputs the corresponding predicted probability p. The myocardial strain parameter vector X consists of six types of left ventricular parameters and three types of left atrial parameters. The six types of left ventricular parameters include GLS, GCS, GRS, cTSD, PSSR, and PDSR parameters. The three types of left atrial parameters include Ee, sRe, and sRa parameters. The predicted probability p is a probability value between 0 and 1.

[0009] A generalized population model dataset is constructed as the corresponding first dataset; and a specified population model dataset is constructed as the corresponding second dataset based on preset specified population characteristics; the first dataset includes multiple first data records; the first data record includes a first training parameter vector and a first label probability; the second dataset includes multiple second data records; the second data record includes a second training parameter vector and a second label probability; both the first and second training parameter vectors are composed of the six types of left ventricular parameters and the three types of left atrial parameters; both the first and second label probabilities are label probabilities with a value of 0 or 1; the specified population characteristics include regional configuration, age group configuration, and gender configuration; when the regional configuration is not empty, it consists of one or more regions; when the age group configuration is not empty, it consists of one or more age groups selected from the preset adolescent age group, middle-aged age group, and elderly age group; the gender configuration includes male, female, and gender-neutral selection;

[0010] The hypertrophic cardiomyopathy risk prediction model is pre-trained based on the first dataset;

[0011] After pre-training, the hypertrophic cardiomyopathy risk prediction model is fine-tuned based on the second dataset;

[0012] After fine-tuning, the set of myocardial strain parameters of the subjects that meet the characteristics of the specified population, input by the user, is used as the corresponding myocardial strain parameter vector X and input into the hypertrophic cardiomyopathy risk prediction model for prediction. The predicted probability p output by this prediction is fed back to the current user. The set of myocardial strain parameters of the subjects is a set of myocardial strain parameters of subjects that meet the characteristics of the specified population, which consists of the six types of left ventricular parameters and the three types of left atrial parameters.

[0013] Preferably, the model input terminal of the hypertrophic cardiomyopathy risk prediction model is used to receive the myocardial strain parameter vector X, and the model output terminal is used to output the corresponding prediction probability p;

[0014] The hypertrophic cardiomyopathy risk prediction model includes a standardization module, a ventricular feature extraction module, an atrial feature extraction module, a feature splicing module, a multi-head self-attention module, a residual connection module, a feature mapping module, and a Sigmoid function module.

[0015] The input of the standardization module is connected to the input of the model, and the first and second outputs are connected to the inputs of the ventricular and atrial feature extraction modules, respectively. The outputs of the ventricular and atrial feature extraction modules are connected to the first and second inputs of the feature splicing module, respectively. The output of the feature splicing module is connected to the input of the multi-head self-attention module and the first input of the residual connection module, respectively. The output of the multi-head self-attention module is connected to the second input of the residual connection module. The output of the residual connection module is connected to the input of the feature mapping module. The output of the feature mapping module is connected to the input of the Sigmoid function module. The output of the Sigmoid function module is connected to the model output.

[0016] The standardization module is used to extract the corresponding six types of left ventricular parameters and three types of left atrial parameters from the myocardial strain parameter vector X; and to standardize the nine types of myocardial strain parameters according to preset standardization rules to obtain the corresponding six types of standardized left ventricular parameters and three types of standardized left atrial parameters; and to form the corresponding left ventricular parameter vector X from the six types of standardized left ventricular parameters. V The left atrial parameter vector X, composed of the three types of standardized left atrial parameters, is sent to the ventricular feature extraction module. A Send to the atrial feature extraction module; wherein, the standardization rules include at least the z-score rule; the six types of left ventricular standardized parameters include GLS standardized parameters, GCS standardized parameters, GRS standardized parameters, cTSD standardized parameters, PSSR standardized parameters, and PDSR standardized parameters; the three types of left atrial standardized parameters include Ee standardized parameters, sRe standardized parameters, and sRa standardized parameters; the left ventricular parameter vector X V It consists of 6 parameters Composition, 1 ≤ index i ≤ 6, 6 parameters Each of the six types of standardized left ventricular parameters corresponds one-to-one; the left atrial parameter vector X A It consists of 3 parameters Composition, 1 ≤ index j ≤ 3, 3 parameters Each of the three types of standardized parameters of the left atrium corresponds one-to-one;

[0017] The ventricular feature extraction module consists of two fully connected layers connected sequentially, with a Leaky ReLU activation function following each fully connected layer; the ventricular feature extraction module is used to process the left ventricular parameter vector X. V Feature extraction is performed to obtain the corresponding feature vector H. V Send to the feature splicing module;

[0018] The feature vector H V The reasoning process is as follows:

[0019] H V =LeakyReLU(W V,2 LeakyReLU(W V,1 X V +b V,1 )+b V,2 ),

[0020] W V,1 b V,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer of the ventricular feature extraction module. V,2 b V,2 The weight matrix parameters and offset vector parameters are those of the second fully connected layer of the ventricular feature extraction module; the left ventricular parameter vector X V The shape is 6×1, and the weight matrix parameter W V,1 The shape is C V,1 ×6, the offset vector parameter b V,1 The shape is C V,1 ×1, LeakyReLU(W V,1 X V +b V,1 The shape of the output vector is C. V,1 ×1; the weight matrix parameter W V,2 The shape is C V,2 ×C V,1 The offset vector parameter b V,2 The shape is C V,2 ×1, the feature vector H V The shape is C V,2 ×1;C V,1 C V,2 These are two preset feature vector dimensions, and 6 < C. V,1 <C V,2 ;

[0021] The atrial feature extraction module consists of two fully connected layers connected sequentially, with a LeakyReLU activation function following each fully connected layer; the atrial feature extraction module is used to process the left atrial parameter vector X. AFeature extraction is performed to obtain the corresponding feature vector H. A Send to the feature splicing module;

[0022] The feature vector H A The reasoning process is as follows:

[0023] H A =LeakyReLU(W A,2 LeakyReLU(W A,1 X A +b A,1 )+b A,2 ),

[0024] W A,1 b A,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer of the atrial feature extraction module. A,2 b A,2 The weight matrix parameters and offset vector parameters are those of the second fully connected layer of the atrial feature extraction module; the left atrial parameter vector X A The shape is 3×1, and the weight matrix parameter W A,1 The shape is C A,1 ×3, the offset vector parameter b A,1 The shape is C A,1 ×1, LeakyReLU(W A,1 X A +b A,1 The shape of the output vector is C. A,1 ×1; the weight matrix parameter W A,2 The shape is C A,2 ×C A,1 The offset vector parameter b A,2 The shape is C A,2 ×1, the feature vector H A The shape is C A,2 ×1;C A,1 C A,2 These are two preset feature vector dimensions, and 3 < C. A,1 <C A,2 C A,1 <C V,1 C A,2 <C V,2 ;

[0025] The feature concatenation module is used to concatenate the feature vector H. V and the feature vector H A The corresponding concatenated vector H is obtained by performing vector concatenation. C The concatenation vector H is sent to the multi-head self-attention module and the residual connection module.C The shape is C C ×1, C C For a predefined feature vector dimension, C C =C V,2 +C A,2 ;

[0026] The multi-head self-attention module consists of N head It consists of N parallel single-head attention units and one multi-head fusion unit; head The preset total number of single heads, N head A feature vector dimension C is greater than 1. C Divisible positive integers; each of the single-head attention units corresponds to a set of independent query / key / value matrix parameters. 1 ≤ index n ≤ N head Each group of query / key / value matrix parameters The shape is C head ×C C C head For a predefined feature vector dimension, C head =C C / N head The multi-head fusion unit corresponds to a fusion matrix parameter W. R The fusion matrix parameter W R The shape is C C ×C C ;

[0027] The multi-head self-attention module is used to process the concatenated vector H. C Perform multi-head self-attention operation to obtain the corresponding feature vector H R Sending to the residual connection module specifically involves each of the single-head attention units using the corresponding query / key / value matrix parameters. For the splicing vector H C Perform a self-attention operation to obtain the corresponding single-head feature vector H. n Send to the multi-head fusion unit; the multi-head fusion unit then processes N. head The single-head feature vector H n Feature concatenation is performed to obtain the corresponding concatenation vector H. all-head And based on the fusion matrix parameter W R For the splicing vector H all-head The corresponding feature vector H is obtained by performing feature mapping processing. R Send to the residual connection module;

[0028] Each of the single-head feature vectors H n The reasoning process is as follows:

[0029]

[0030] Q n K n V n Let the query vector, key vector, and value vector be the nth single-head attention unit. Let K be the key vector n The transpose of the vector, softmax() is the Softmax function; the query, key, and value vector Q n K n V n The shape is C head ×1, the transpose vector The shape is 1×C head The single-head feature vector H n The shape is C head ×1;

[0031] The splicing vector H all-head The shape is (C head ×N head )×1=C C ×1;

[0032] The feature vector H R The reasoning process is as follows:

[0033] H R =W R H all-head ,

[0034] The feature vector H R The shape is C C ×1;

[0035] The residual connection module is used to process the splicing vector H according to the residual connection method. C and the feature vector H R The corresponding feature vector H is obtained by fusion. R+C Send to the feature mapping module;

[0036] The feature vector H R+C The reasoning process is as follows:

[0037] H R+c =H c +H R ,

[0038] The feature vector H R+C The shape is C C ×1;

[0039] The feature mapping module consists of two fully connected layers connected sequentially, with a Swish activation function following the first fully connected layer; the feature mapping module is used to map the feature vector H... R+C The confidence score is calculated and the corresponding confidence score z is sent to the Sigmoid function module;

[0040] The reasoning process for the confidence score z is as follows:

[0041] z = W out,2 Swish(W out,1 H R+C +b out,1 )+b out,2 ,

[0042] W out,1 b out,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer of the feature mapping module. out,2 b out,2 The weight matrix parameters and offset vector parameters are those of the second fully connected layer of the feature mapping module; the weight matrix parameter W out,1 The shape is C out,1 ×C C The offset vector parameter b V,1 The shape is C out,1 ×1, Swish(W out,1 H R+C +b out,1 The shape of the output vector is C. out,1 ×1; the weight matrix parameter W out,2 The shape is C out,2 ×C out,1 The offset vector parameter b out,2 The shape is C out,2 ×1, C out,1 C out,2 These are two preset feature vector dimensions, C. out,1 >C C C out,2 =1; the confidence score z is a scalar;

[0043] The Sigmoid function module is used to input the confidence score z into the Sigmoid function for calculation and output the calculation result as the corresponding prediction probability p.

[0044] Preferably, the construction of the generalized population model dataset, as the corresponding first dataset, specifically includes:

[0045] Step 31: Collect large data of cardiac magnetic resonance images of multiple populations of multiple ages in multiple regions through preset multiple data collection channels to obtain the corresponding original image set.

[0046] The data collection channels include at least publicly available medical image datasets, publicly available or authorized medical institution databases, and publicly available or authorized medical image sharing platforms; the regions include at least two or more regions; the age groups include at least adolescents, middle-aged individuals, and the elderly; the population groups include at least healthy individuals without any heart disease, other patients without hypertrophic cardiomyopathy but with other heart diseases, and a target patient population with hypertrophic cardiomyopathy; the gender of the population groups should cover both male and female; the original image set includes multiple original image records; each original image record includes a first image, a first region, a first gender, a first age group, and a first population type; the first image is a cardiac magnetic resonance imaging (MRI) image; the first region is one of the multiple regions; the first gender includes males and females; the first age group includes adolescents, middle-aged individuals, and the elderly; the first population type includes healthy individuals without any heart disease, other patients without hypertrophic cardiomyopathy but with other heart diseases, and a target patient population with hypertrophic cardiomyopathy.

[0047] Step 32: Take the first image of each original image record in the original image set as the corresponding current image; and quantitatively evaluate the six types of left ventricular parameters and the three types of left atrial parameters corresponding to the current image based on a preset myocardial strain parameter evaluation software, and form a corresponding first training parameter vector from the evaluated six types of left ventricular parameters and the three types of left atrial parameters; identify whether the first population type corresponding to the current image is a target patient population with hypertrophic cardiomyopathy, and if so, set a corresponding first label probability of 1, otherwise set a corresponding first label probability of 0; and form a corresponding first data record from the first training parameter vector corresponding to the current image and the first label probability; and form a corresponding first dataset from all the obtained first data records.

[0048] The myocardial strain parameter assessment software includes at least CVI42 software, CAAS MR Strain software, and Segment software.

[0049] Preferably, the step of constructing a specified population model dataset based on preset specified population characteristics as the corresponding second dataset specifically includes:

[0050] Extract the corresponding region configuration, age group configuration, and gender configuration from the specified population characteristics;

[0051] The system identifies whether the region configuration is empty; if so, the original image set is used as the corresponding first image set; otherwise, the original image records in the original image set that satisfy the region configuration in the first region are used to form the corresponding first image set.

[0052] The system identifies whether the age group configuration is empty; if so, the first image set is used as the corresponding second image set; if not, the original image records in the first image set that satisfy the age group configuration are used to form the corresponding second image set.

[0053] The gender configuration is identified; if the gender configuration is male, the corresponding third image set is composed of the original image records of the first gender being male in the second image set; if the gender configuration is female, the corresponding third image set is composed of the original image records of the first gender being female in the second image set; if the gender configuration is both male and female, the second image set is used as the corresponding third image set.

[0054] The first data record corresponding to each original image record in the third image set is extracted from the first dataset and used as a corresponding second data record; and all the obtained second data records are used to form the corresponding second dataset.

[0055] Preferably, the pre-training of the hypertrophic cardiomyopathy risk prediction model based on the first dataset specifically includes:

[0056] Step 51: Based on the preset first segmentation ratio, the first dataset is randomly segmented into a first training set and a first evaluation set; the full model parameter set of the hypertrophic cardiomyopathy risk prediction model is used as the corresponding current model parameter set; and the first training set is divided into multiple first batch record sets according to the preset maximum number of records per batch.

[0057] Wherein, both the first training set and the first evaluation set consist of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio; the total number of first data records in the first batch of record sets is less than or equal to the maximum number of records in a single batch;

[0058] Step 52: Take the first batch of records as the corresponding current record set;

[0059] Step 53: Perform a traversal of all the first data records in the current record set; during this traversal, the first data record currently traversed is used as the corresponding current training record; the first training parameter vector of the current training record is used as the current myocardial strain parameter vector X and input into the hypertrophic cardiomyopathy risk prediction model for prediction, and the prediction probability p output by this prediction is used as the corresponding first prediction probability; the current first prediction probability and the first label probability of the current training record form a corresponding first prediction-label pair; at the end of this traversal, all the first prediction-label pairs obtained in this traversal are substituted into a preset first model loss function to calculate the corresponding first loss value;

[0060] The first model loss function is implemented based on the L1 loss function, the L2 loss function, or the binary cross-entropy loss function.

[0061] Step 54: Identify whether the first loss value meets the preset first loss value range; if the first loss value meets the first loss value range, identify whether the current record set is the last of the first batch of record sets in the first training set; if so, proceed to step 55; otherwise, take the next batch of record sets as the new current record set and return to step 53; if the first loss value does not meet the first loss value range, perform a round of modulation on the current model parameter set based on the preset first model optimizer in the direction of minimizing the first model loss function, and return to step 53 after the end of this round of modulation.

[0062] The first model optimizer includes at least the AdamW optimizer, the Adam optimizer, and the SGD optimizer.

[0063] Step 55: Perform a traversal of all the first data records in the first evaluation set; during this traversal, take the currently traversed first data record as the corresponding current evaluation record; and input the first training parameter vector of the current evaluation record as the current myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and take the prediction probability p output by this prediction as the corresponding second prediction probability; and form a corresponding second prediction-label pair with the current second prediction probability and the first label probability of the current evaluation record; and at the end of this traversal, input all the obtained second prediction-label pairs into the preset first model evaluation function to calculate the corresponding first evaluation value;

[0064] The first model evaluation function is implemented based on the MAE function, MSE function, or RMSE function.

[0065] Step 56: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 52; if it does, confirm that the model pre-training is complete.

[0066] Preferably, the fine-tuning of the hypertrophic cardiomyopathy risk prediction model based on the second dataset specifically includes:

[0067] Step 61: Based on a preset second segmentation ratio, the second dataset is randomly subdivided into corresponding second training sets and second evaluation sets; and the fusion matrix parameters W of the multi-head self-attention module of the hypertrophic cardiomyopathy risk prediction model are used. R and the weight matrix parameter W of the feature mapping module out,1 W out,2 The corresponding current fine-tuning parameter set is formed; and the second training set is divided into multiple second batch record sets according to the preset maximum number of records per batch;

[0068] The second training set and the second evaluation set are each composed of multiple second data records; the ratio of the total number of records in the second training set to the total number of records in the second evaluation set satisfies the second partitioning ratio; the total number of second data records in the second batch of record sets is less than or equal to the maximum number of records in a single batch.

[0069] Step 62: Take the first set of records from the second batch as the corresponding current set;

[0070] Step 63: Perform a traversal of all the second data records in the current record set; during this traversal, the currently traversed second data record is used as the corresponding current training record; the second training parameter vector of the current training record is used as the current myocardial strain parameter vector X and input into the hypertrophic cardiomyopathy risk prediction model for prediction, and the prediction probability p output by this prediction is used as the corresponding third prediction probability; the current third prediction probability and the second label probability of the current training record form a corresponding third prediction-label pair; at the end of this traversal, all the third prediction-label pairs obtained in this traversal are substituted into the preset second model loss function to calculate the corresponding second loss value;

[0071] The second model loss function is based on the binary cross-entropy loss function.

[0072] Step 64: Identify whether the second loss value meets the preset second loss value range; if the second loss value meets the second loss value range, identify whether the current record set is the last second batch of record sets in the second training set; if so, proceed to step 65; otherwise, take the next second batch of record sets as the new current record set and return to step 63; if the second loss value does not meet the second loss value range, perform a round of fine-tuning on the current model parameter set based on the preset second model optimizer in the direction of minimizing the second model loss function, and return to step 63 after this round of fine-tuning.

[0073] The second model optimizer includes at least the AdamW optimizer, the Adam optimizer, and the SGD optimizer;

[0074] Step 65: Perform a round of traversal on all the second data records in the second evaluation set; during this round of traversal, take the currently traversed second data record as the corresponding current evaluation record; and input the second training parameter vector of the current evaluation record as the current myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and take the prediction probability p output by this prediction as the corresponding fourth prediction probability; and form a corresponding fourth prediction-label pair by the current fourth prediction probability and the second label probability of the current evaluation record; and at the end of this round of traversal, calculate the corresponding first accuracy, first precision, first recall, and first F1 score based on all the obtained fourth prediction-label pairs.

[0075] Step 66: Identify whether the first accuracy, the first precision, the first recall, and the first F1 score each satisfy the corresponding first accuracy range, first precision range, first recall range, and first F1 score range; if not, return to step 62; if yes, confirm that the model fine-tuning is complete.

[0076] A second aspect of the present invention provides an apparatus for implementing the processing method of the hypertrophic cardiomyopathy risk prediction model described in the first aspect above, the apparatus comprising: a model building module, a dataset building module, a model pre-training module, a model fine-tuning module, and a model application module;

[0077] The model building module is used to construct a deep learning model for predicting the risk probability of hypertrophic cardiomyopathy, denoted as the corresponding hypertrophic cardiomyopathy risk prediction model. The hypertrophic cardiomyopathy risk prediction model predicts the risk probability of hypertrophic cardiomyopathy based on the myocardial strain parameter vector X input to the model and outputs the corresponding predicted probability p. The myocardial strain parameter vector X consists of six types of left ventricular parameters and three types of left atrial parameters. The six types of left ventricular parameters include GLS parameters, GCS parameters, GRS parameters, cTSD parameters, PSSR parameters, and PDSR parameters. The three types of left atrial parameters include Ee parameters, sRe parameters, and sRa parameters. The predicted probability p is a probability value between 0 and 1.

[0078] The dataset construction module is used to construct a generalized population model dataset as the corresponding first dataset; and to construct a specified population model dataset as the corresponding second dataset based on preset specified population characteristics; the first dataset includes multiple first data records; the first data record includes a first training parameter vector and a first label probability; the second dataset includes multiple second data records; the second data record includes a second training parameter vector and a second label probability; the first and second training parameter vectors are both composed of the six types of left ventricular parameters and the three types of left atrial parameters; the first and second label probabilities are both label probabilities with a value of 0 or 1; the specified population characteristics include regional configuration, age group configuration, and gender configuration; when the regional configuration is not empty, it consists of one or more regions; when the age group configuration is not empty, it consists of one or more age groups among the preset adolescent age group, middle-aged age group, and elderly age group; the gender configuration includes male, female, and gender-neutral selection;

[0079] The model pre-training module is used to pre-train the hypertrophic cardiomyopathy risk prediction model based on the first dataset;

[0080] The model fine-tuning module is used to fine-tune the hypertrophic cardiomyopathy risk prediction model based on the second dataset after pre-training.

[0081] The model application module is used to, after fine-tuning, input the set of myocardial strain parameters of the subjects that meet the characteristics of the specified population as the corresponding myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and feed back the prediction probability p output by this prediction to the current user; the set of myocardial strain parameters of the subjects is a set of myocardial strain parameters of subjects that meet the characteristics of the specified population, which consists of the six types of left ventricular parameters and the three types of left atrial parameters.

[0082] A third aspect of the present invention provides an electronic device, including: a memory, a processor, and a transceiver;

[0083] The processor is used to couple with the memory, read and execute instructions in the memory to implement the steps of the method described in the first aspect above;

[0084] The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

[0085] A fourth aspect of the present invention provides a computer-readable storage medium storing computer instructions that, when executed by a computer, cause the computer to perform the instructions described in the first aspect.

[0086] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for processing a hypertrophic cardiomyopathy (HCM) risk prediction model. As described above, this invention constructs a HCM risk prediction model capable of predicting risk based on six types of left ventricular parameters (GLS, GCS, GRS, cTSD, PSSR, and PDSR) and three types of left atrial parameters (Ee, sRe, and sRa). It also constructs a generalized population model dataset (first dataset) and a specified population model dataset (second dataset) through big data collection. The HCM risk prediction model is first pre-trained based on the first dataset and then fine-tuned based on the second dataset. After fine-tuning, the HCM risk prediction model is used to predict the risk based on the myocardial strain parameters of the subjects. This invention improves both prediction accuracy and prediction stability. Attached Figure Description

[0087] Figure 1 This is a schematic diagram of a processing method for a hypertrophic cardiomyopathy risk prediction model provided in Embodiment 1 of the present invention;

[0088] Figure 2 A schematic diagram of the module of the hypertrophic cardiomyopathy risk prediction model provided in Embodiment 1 of the present invention;

[0089] Figure 3 This is a module structure diagram of a processing device for a hypertrophic cardiomyopathy risk prediction model provided in Embodiment 2 of the present invention;

[0090] Figure 4 This is a schematic diagram of the structure of an electronic device provided in Embodiment 3 of the present invention. Detailed Implementation

[0091] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0092] Embodiment 1 of the present invention provides a method for processing a risk prediction model for hypertrophic cardiomyopathy, such as... Figure 1 The schematic diagram shows a processing method for a hypertrophic cardiomyopathy risk prediction model provided in Embodiment 1 of the present invention. The method mainly includes the following steps:

[0093] Step 1: Construct a deep learning model for predicting the risk probability of hypertrophic cardiomyopathy, denoted as the corresponding hypertrophic cardiomyopathy risk prediction model.

[0094] Here, the hypertrophic cardiomyopathy risk prediction model of this invention is used to predict the risk probability of hypertrophic cardiomyopathy based on the myocardial strain parameter vector X input to the model and output the corresponding predicted probability p; for example Figure 2 The diagram shows a module schematic of the hypertrophic cardiomyopathy risk prediction model provided in Embodiment 1 of the present invention. The myocardial strain parameter vector X consists of six types of left ventricular parameters and three types of left atrial parameters. The six types of left ventricular parameters include GLS, GCS, GRS, cTSD, PSSR, and PDSR parameters; the three types of left atrial parameters include Ee, sRe, and sRa parameters. The prediction probability p is a probability value between 0 and 1.

[0095] like Figure 2 As shown, the model input of the hypertrophic cardiomyopathy risk prediction model is used to receive the myocardial strain parameter vector X, and the model output is used to output the corresponding prediction probability p.

[0096] The model components of the hypertrophic cardiomyopathy risk prediction model include: a standardization module, a ventricular feature extraction module, an atrial feature extraction module, a feature splicing module, a multi-head self-attention module, a residual connection module, a confidence assessment module, and a sigmoid function module.

[0097] The connection relationships of the model components in the hypertrophic cardiomyopathy risk prediction model are as follows: the input of the standardization module is connected to the model input; the first and second outputs are connected to the inputs of the ventricular and atrial feature extraction modules, respectively; the outputs of the ventricular and atrial feature extraction modules are connected to the first and second inputs of the feature splicing module, respectively; the output of the feature splicing module is connected to the input of the multi-head self-attention module and the first input of the residual connection module, respectively; the output of the multi-head self-attention module is connected to the second input of the residual connection module; the output of the residual connection module is connected to the input of the confidence assessment module; the output of the confidence assessment module is connected to the input of the Sigmoid function module; and the output of the Sigmoid function module is connected to the model output.

[0098] The model components of the hypertrophic cardiomyopathy risk prediction model are shown below.

[0099] 1) Standardized Module:

[0100] The standardization module in this embodiment of the invention is used to extract six types of left ventricular parameters and three types of left atrial parameters from the myocardial strain parameter vector X; and to standardize the nine types of myocardial strain parameters according to preset standardization rules to obtain the corresponding six types of standardized left ventricular parameters and three types of standardized left atrial parameters; and to form the corresponding left ventricular parameter vector X from the six types of standardized left ventricular parameters. V The left atrial parameter vector X, composed of three types of standardized left atrial parameters, is sent to the ventricular feature extraction module. A Send to the atrial feature extraction module.

[0101] The standardization rules include at least the z-score rule; six types of standardized left ventricular parameters include GLS, GCS, GRS, cTSD, PSSR, and PDSR parameters; three types of standardized left atrial parameters include Ee, sRe, and sRa parameters; and the left ventricular parameter vector X... V It consists of 6 parameters Composition, 1 ≤ index i ≤ 6, 6 parameters Each parameter corresponds one-to-one with one of the six types of standardized left ventricular parameters; the left atrial parameter vector X A It consists of 3 parameters Composition, 1 ≤ index j ≤ 3, 3 parameters It corresponds one-to-one with the three types of standardized parameters of the left atrium.

[0102] 2) Ventricular Feature Extraction Module:

[0103] The ventricular feature extraction module of this invention consists of two fully connected layers connected sequentially, with a LeakyReLU activation function following each fully connected layer. The ventricular feature extraction module is used to process the left ventricular parameter vector X. V Feature extraction is performed to obtain the corresponding feature vector H. V Send to the feature splicing module.

[0104] Here, the feature vector H V The reasoning process is as follows:

[0105] H V =LeakyReLU(W V,2 LeakyReLU(W V,1 X V +b V,1 )+b V,2 );

[0106] Among them, W V,1 b V,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer of the ventricular feature extraction module. V,2 b V,2 The parameters are the weight matrix and offset vector of the second fully connected layer in the ventricular feature extraction module; the left ventricular parameter vector X. V The shape is 6×1, and the weight matrix parameter W V,1 The shape is C V,1 ×6, offset vector parameter b V,1 The shape is C V,1 ×1, LeakyReLU(W V,1 X V +b V,1 The shape of the output vector is C. V,1 ×1; Weight matrix parameter W V,2 The shape is C V,2 ×C V,1 offset vector parameter b V,2 The shape is C V,2 ×1, eigenvector H V The shape is C V,2 ×1;C V,1 C V,2 These are two preset feature vector dimensions, and 6 < C. V,1 <C V,2 For example, C V,1 =12, C V,2 =16.

[0107] 3) Atrial Feature Extraction Module:

[0108] The atrial feature extraction module of this invention consists of two fully connected layers connected sequentially, with a LeakyReLU activation function following each fully connected layer. The atrial feature extraction module is used to process the left atrial parameter vector X. A Feature extraction is performed to obtain the corresponding feature vector H. A Send to the feature splicing module.

[0109] Here, the feature vector H A The reasoning process is as follows:

[0110] H A =LeakyReLU(W A,2 LeakyReLU(W A,1 X A +b A,1 )+b A,2 );

[0111] Among them, W A,1 b A,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer in the atrial feature extraction module. A,2 b A,2 The parameters are the weight matrix and offset vector of the second fully connected layer in the atrial feature extraction module; the left atrial parameter vector X. A The shape is 3×1, and the weight matrix parameter W A,1 The shape is C A,1 ×3, offset vector parameter b A,1 The shape is C A,1 ×1, LeakyReLU(W A,1 X A +b A,1 The shape of the output vector is C. A,1 ×1; Weight matrix parameter W A,2 The shape is C A,2 ×C A,1 offset vector parameter b A,2 The shape is C A,2 ×1, eigenvector H A The shape is C A,2 ×1;C A,1 C A,2 These are two preset feature vector dimensions, and 3 < C. A,1 <C A,2 C A,1 <C V,1 C A,2 <C V,2 For example, C A,1 =6, C A,2 =8.

[0112] 4) Feature splicing module:

[0113] The feature concatenation module in this embodiment of the invention is used to concatenate feature vector H V and eigenvector H A The corresponding concatenated vector H is obtained by performing vector concatenation. C Send to the multi-head self-attention module and the residual connection module.

[0114] Here, the concatenated vector H C The shape is C C ×1, C C For a predefined feature vector dimension, C C =C V,2 +C A,2 .

[0115] 5) Multi-head self-attention module:

[0116] The multi-head self-attention module of this invention consists of N head It consists of N parallel single-head attention units and one multi-head fusion unit; where N head The preset total number of single heads, N. head A value greater than 1 that can be C C Divisible positive integers, such as N head =4; Each single-head attention unit corresponds to a set of independent query / key / value matrix parameters. 1 ≤ index n ≤ N head Each set of query / key / value matrix parameters The shape is C head ×C C C head For a predefined feature vector dimension, C head =C C / N head A multi-head fusion unit corresponds to a fusion matrix parameter W. R fusion matrix parameter W R The shape is C C ×C C .

[0117] The multi-head self-attention module is used to concatenate vector H C Perform multi-head self-attention operation to obtain the corresponding feature vector H R Sending to the residual connection module, specifically: each single-head attention unit based on the corresponding query / key / value matrix parameters. For the concatenated vector H C Perform a self-attention operation to obtain the corresponding single-head feature vector H. n Send to the multi-head fusion unit; the multi-head fusion unit then processes N. head A single-head feature vector H nFeature concatenation is performed to obtain the corresponding concatenation vector H. all-head And based on the fusion matrix parameter W R For the concatenated vector H all-head The corresponding feature vector H is obtained by performing feature mapping. R Send to the residual connection module.

[0118] Here, each single-head feature vector H n The reasoning process is as follows:

[0119]

[0120] Among them, Q n K n V n Let n be the query vector, key vector, and value vector of the nth single-head attention unit. For the key vector K n The transpose of the vector, softmax() is the Softmax function; query, key, and value vector Q n K n V n The shape is C head ×1, The shape is 1×C head Single-head eigenvector H n The shape is C head ×1.

[0121] Additionally, the concatenated vector H all-head The shape is (C head ×N head )×1=C C ×1.

[0122] In addition, the eigenvector H R The reasoning process is as follows:

[0123] H R =W R H all-head ;

[0124] Among them, the feature vector H R The shape is C C ×1.

[0125] 6) Residual connection module:

[0126] The residual connection module in this embodiment of the invention is used to perform splicing on the spliced ​​vector H according to the residual connection method. C and eigenvector H R The corresponding feature vector H is obtained by fusion. R+C Send to the confidence assessment module.

[0127] Here, the feature vector H R+C The reasoning process is as follows:

[0128] H R+c =H c +H R ;

[0129] Among them, the feature vector H R+C The shape is C C ×1.

[0130] 7) Confidence assessment module:

[0131] The confidence evaluation module in this embodiment of the invention consists of two fully connected layers connected sequentially, with a Swish activation function following the first fully connected layer. The confidence evaluation module is used to evaluate the feature vector H... R+C The confidence score is calculated and the corresponding confidence score z is sent to the Sigmoid function module.

[0132] Here, the reasoning process for the confidence score z is as follows:

[0133] z = W out,2 Swish(W out,1 H R+c +b out,1 )+b out,2 ;

[0134] Among them, W out,1 b out,1 For the weight matrix parameters and offset vector parameters of the first fully connected layer of the confidence evaluation module, W out,2 b out,2 The weight matrix parameters and offset vector parameters are for the second fully connected layer of the confidence evaluation module; the weight matrix parameter W out,1 The shape is C out,1 ×C C offset vector parameter b V,1 The shape is C out,1 ×1, Swish(W out,1 H R+C +b out,1 The shape of the output vector is C. out,1 ×1; Weight matrix parameter W out,2 The shape is C out,2 ×C out,1 offset vector parameter b out,2 The shape is C out,2 ×1, C out,1 C out,2 These are two preset feature vector dimensions, C. out,1 >C C For example, C out,1=32; Furthermore, since the confidence score z is a scalar, the feature vector dimension C out,2 =1.

[0135] 8) Sigmoid function module:

[0136] The Sigmoid function module in this embodiment of the invention is used to input the confidence score z into the Sigmoid function for calculation and output the calculation result as the corresponding prediction probability p.

[0137] It's important to note that the confidence score z can be considered an assessment score used to evaluate the confidence level of the current sample (i.e., the current model input myocardial strain parameter vector X) as a positive sample (i.e., the current myocardial strain parameter vector X is the myocardial strain parameter vector of a patient with hypertrophic cardiomyopathy). A higher confidence score z means a higher probability that the current sample is a positive sample, and vice versa. A lower confidence score z means a lower probability that the current sample is a positive sample and a higher probability that it is a negative sample. The prediction probability p, converted from the confidence score z, can be considered the probability that the current sample is a positive sample. A higher prediction probability p means a higher probability that the current sample is a positive sample, and vice versa.

[0138] Step 2: Construct a generalized population model dataset as the corresponding first dataset; and construct a specified population model dataset based on the preset characteristics of the specified population as the corresponding second dataset;

[0139] The first dataset includes multiple first data records; each first data record includes a first training parameter vector and a first label probability; the second dataset includes multiple second data records; each second data record includes a second training parameter vector and a second label probability; both the first and second training parameter vectors consist of six classes of left ventricular parameters and three classes of left atrial parameters; both the first and second label probabilities are label probabilities that take a value of 0 or 1.

[0140] The specified demographic characteristics include region configuration, age group configuration, and gender configuration; when the region configuration is not empty, it consists of one or more regions; when the age group configuration is not empty, it consists of one or more age groups from the preset youth age group, middle-aged age group, and elderly age group; the gender configuration includes male, female, and both genders.

[0141] The current step 2 specifically includes:

[0142] Step 21: Construct a generalized population model dataset as the corresponding first dataset;

[0143] Specifically, it includes: Step 211, collecting cardiac magnetic resonance images of multiple populations in multiple regions and age groups through preset multiple data collection channels to obtain the corresponding original image set;

[0144] Here, the various data acquisition channels in this embodiment of the invention include at least publicly available medical image datasets, publicly available or authorized medical institution databases, and publicly available or authorized medical image sharing platforms.

[0145] The multiple regions in this embodiment of the invention include at least two or more regions;

[0146] The multiple age groups in the embodiments of the present invention include at least the adolescent age group, the middle-aged age group, and the elderly age group;

[0147] The various population groups in this invention include at least healthy individuals without any heart disease, other patients without hypertrophic cardiomyopathy but with other heart diseases, and target patient populations with hypertrophic cardiomyopathy.

[0148] The gender of the various population groups in this invention should cover both male and female.

[0149] The original image set in this embodiment of the invention includes multiple original image records; each original image record includes a first image, a first region, a first gender, a first age group, and a first population type; the first image is a cardiac magnetic resonance image; the first region is one of multiple regions; the first gender includes males and females; the first age group includes adolescents, middle-aged individuals, and elderly individuals; the first population type includes healthy individuals without any heart disease, other patients without hypertrophic cardiomyopathy but with other heart diseases, and a target patient population with hypertrophic cardiomyopathy.

[0150] Step 212: The first image of each original image record in the original image set is taken as the corresponding current image; and the six types of left ventricular parameters and three types of left atrial parameters corresponding to the current image are quantitatively evaluated based on the preset myocardial strain parameter evaluation software, and a corresponding first training parameter vector is formed by the evaluated six types of left ventricular parameters and three types of left atrial parameters; and whether the first population type corresponding to the current image is the target patient population with hypertrophic cardiomyopathy is identified. If so, a corresponding first label probability of 1 is set; otherwise, a corresponding first label probability of 0 is set; and a corresponding first data record is formed by the first training parameter vector and the first label probability corresponding to the current image; and a corresponding first dataset is formed by all the obtained first data records.

[0151] Here, the myocardial strain parameter assessment software of the present invention includes at least CVI42 software, CAAS MRStrain software, and Segment software;

[0152] Step 22: Construct a specified population model dataset based on the preset characteristics of the specified population as the corresponding second dataset;

[0153] Specifically, this includes: Step 221, extracting the corresponding region configuration, age group configuration, and gender configuration from the characteristics of the specified population;

[0154] Step 222, and identify whether the region configuration is empty; if so, the original image set is used as the corresponding first image set; if not, the original image records of the first region in the original image set that meet the region configuration are used to form the corresponding first image set.

[0155] Step 223, and identify whether the age group configuration is empty; if so, the first image set is used as the corresponding second image set; if not, the original image records of the first age group in the first image set that meet the age group configuration are used to form the corresponding second image set.

[0156] Step 224, and identify the gender configuration; if the gender configuration is male, the corresponding third image set is composed of the original image records of the first gender being male in the second image set; if the gender configuration is female, the corresponding third image set is composed of the original image records of the first gender being female in the second image set; if the gender configuration is both male and female, the second image set is used as the corresponding third image set.

[0157] Step 225: Extract the first data record corresponding to each original image record in the third image set from the first dataset as a corresponding second data record; and form the corresponding second dataset from all the obtained second data records.

[0158] Step 3: Pre-train the hypertrophic cardiomyopathy risk prediction model based on the first dataset;

[0159] Specifically, it includes: Step 31, randomly subdividing the first dataset into corresponding first training set and first evaluation set based on a preset first segmentation ratio; taking the full model parameter set of the hypertrophic cardiomyopathy risk prediction model as the corresponding current model parameter set; and dividing the first training set into multiple first batch record sets according to the preset maximum number of records per batch.

[0160] Wherein, the first split ratio is a pre-set ratio parameter, such as 8:2; both the first training set and the first evaluation set consist of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first split ratio; the maximum number of records in a single batch is a pre-set positive integer, such as 10; the total number of first data records in the first batch of record sets is less than or equal to the maximum number of records in a single batch.

[0161] Step 32: Take the first batch of records as the corresponding current record set;

[0162] Step 33: Perform a round of traversal on all first data records in the current record set; during this round of traversal, use the first data record currently traversed as the corresponding current training record; and use the first training parameter vector of the current training record as the current myocardial strain parameter vector X to input into the hypertrophic cardiomyopathy risk prediction model for prediction, and use the prediction probability p output by this prediction as the corresponding first prediction probability; and form a corresponding first prediction-label pair by the current first prediction probability and the first label probability of the current training record; and at the end of this round of traversal, substitute all the first prediction-label pairs obtained in this round of traversal into the preset first model loss function to calculate the corresponding first loss value;

[0163] The first model loss function is implemented based on the L1 loss function, L2 loss function, or binary cross-entropy loss function.

[0164] Step 34: Identify whether the first loss value meets the preset first loss value range; if the first loss value meets the first loss value range, identify whether the current record set is the last batch of record sets in the first training set; if so, proceed to step 35; otherwise, take the next batch of record sets as the new current record set and return to step 33; if the first loss value does not meet the first loss value range, perform a round of modulation on the current model parameter set based on the preset first model optimizer in the direction of minimizing the first model loss function, and return to step 33 after the end of this round of modulation.

[0165] The first loss value range is a pre-set numerical range; the first model optimizer includes at least the AdamW optimizer, the Adam optimizer, and the SGD optimizer.

[0166] Step 35: Perform a round of traversal on all first data records in the first evaluation set; during this round of traversal, take the currently traversed first data record as the corresponding current evaluation record; and input the first training parameter vector of the current evaluation record as the current myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and take the prediction probability p output by this prediction as the corresponding second prediction probability; and form a corresponding second prediction-label pair by the current second prediction probability and the first label probability of the current evaluation record; and at the end of this round of traversal, input all the obtained second prediction-label pairs into the preset first model evaluation function to calculate the corresponding first evaluation value;

[0167] The first model evaluation function is implemented based on the MAE function, MSE function, or RMSE function.

[0168] Step 36: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 32; if it does, confirm that the model pre-training is complete.

[0169] Here, the first evaluation value range is a pre-set numerical range.

[0170] Step 4: After pre-training, fine-tune the hypertrophic cardiomyopathy risk prediction model based on the second dataset;

[0171] Specifically, this includes: Step 41, randomly subdividing the second dataset into corresponding second training sets and second evaluation sets based on a preset second segmentation ratio; and using the fusion matrix parameters W of the multi-head self-attention module of the hypertrophic cardiomyopathy risk prediction model. R And the weight matrix parameter W of the confidence assessment module out,1 W out,2 The corresponding current fine-tuning parameter set is formed; and the second training set is divided into multiple second batch record sets according to the preset maximum number of records per batch;

[0172] The second split ratio is a pre-set ratio parameter, such as 8:2; both the second training set and the second evaluation set consist of multiple second data records; the ratio of the total number of records in the second training set to the total number of records in the second evaluation set satisfies the second split ratio; the total number of second data records in the second batch of records is less than or equal to the maximum number of records in a single batch.

[0173] Step 42: Use the first second batch of record sets as the corresponding current record set;

[0174] Step 43: Perform a round of traversal on all second data records in the current record set; during this round of traversal, use the second data record currently traversed as the corresponding current training record; input the second training parameter vector of the current training record as the current myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and use the prediction probability p output by this prediction as the corresponding third prediction probability; form a corresponding third prediction-label pair with the second label probability of the current training record; and at the end of this round of traversal, substitute all the third prediction-label pairs obtained in this round of traversal into the preset second model loss function to calculate the corresponding second loss value;

[0175] The second model loss function is based on the binary cross-entropy loss function.

[0176] Step 44: Identify whether the second loss value meets the preset range of the second loss value; if the second loss value meets the range of the second loss value, identify whether the current record set is the last second batch of record sets in the second training set; if so, proceed to step 45; otherwise, take the next second batch of record sets as the new current record set and return to step 43; if the second loss value does not meet the range of the second loss value, perform a round of fine-tuning on the current model parameter set based on the preset second model optimizer in the direction of minimizing the second model loss function, and return to step 43 after the fine-tuning is completed.

[0177] The second loss value range is a pre-set numerical range; the second model optimizer includes at least the AdamW optimizer, the Adam optimizer, and the SGD optimizer.

[0178] Step 45: Perform a round of traversal on all second data records in the second evaluation set; during this round of traversal, the currently traversed second data record is taken as the corresponding current evaluation record; the second training parameter vector of the current evaluation record is taken as the current myocardial strain parameter vector X and input into the hypertrophic cardiomyopathy risk prediction model for prediction, and the prediction probability p output by this prediction is taken as the corresponding fourth prediction probability; and the current fourth prediction probability and the second label probability of the current evaluation record form a corresponding fourth prediction-label pair; at the end of this round of traversal, the accuracy, precision, recall and F1 score are calculated based on all the obtained fourth prediction-label pairs to obtain the corresponding first accuracy, first precision, first recall and first F1 score;

[0179] Step 46: Identify whether the first accuracy, first precision, first recall, and first F1 score each satisfy the corresponding first accuracy range, first precision range, first recall range, and first F1 score range; if not, return to step 42; if yes, confirm that the model fine-tuning is complete.

[0180] Here, the first accuracy range, the first precision range, the first recall range, and the first F1 score range are four pre-set numerical ranges.

[0181] Step 5: After fine-tuning, the set of myocardial strain parameters of the subjects that meet the characteristics of the specified population, input by the user, is used as the corresponding myocardial strain parameter vector X. This vector is then input into the hypertrophic cardiomyopathy risk prediction model for prediction, and the predicted probability p output by this prediction is fed back to the current user.

[0182] Here, the myocardial strain parameter set of the subjects in this invention is a myocardial strain parameter set of subjects that meets the characteristics of a specified population, consisting of six types of left ventricular parameters and three types of left atrial parameters.

[0183] Figure 3 This is a module structure diagram of a processing device for a hypertrophic cardiomyopathy risk prediction model provided in Embodiment 2 of the present invention. This device can be a terminal device or server implementing the aforementioned method embodiment, or it can be a device that enables the aforementioned terminal device or server to implement the aforementioned method embodiment. For example, the device can be a device or chip system of the aforementioned terminal device or server. Figure 3 As shown, the device includes: a model building module 201, a dataset building module 202, a model pre-training module 203, a model fine-tuning module 204, and a model application module 205.

[0184] The model building module 201 is used to build a deep learning model for predicting the risk probability of hypertrophic cardiomyopathy, denoted as the corresponding hypertrophic cardiomyopathy risk prediction model. The hypertrophic cardiomyopathy risk prediction model is used to predict the risk probability of hypertrophic cardiomyopathy based on the myocardial strain parameter vector X input to the model and output the corresponding prediction probability p. The myocardial strain parameter vector X consists of six types of left ventricular parameters and three types of left atrial parameters. The six types of left ventricular parameters include GLS parameters, GCS parameters, GRS parameters, cTSD parameters, PSSR parameters, and PDSR parameters. The three types of left atrial parameters include Ee parameters, sRe parameters, and sRa parameters. The prediction probability p is a probability value between 0 and 1.

[0185] The dataset construction module 202 is used to construct a generalized population model dataset as the corresponding first dataset; and to construct a specified population model dataset as the corresponding second dataset based on preset specified population characteristics; the first dataset includes multiple first data records; the first data records include a first training parameter vector and a first label probability; the second dataset includes multiple second data records; the second data records include a second training parameter vector and a second label probability; both the first and second training parameter vectors consist of six types of left ventricular parameters and three types of left atrial parameters; both the first and second label probabilities are label probabilities with a value of 0 or 1; the specified population characteristics include regional configuration, age group configuration, and gender configuration; when the regional configuration is not empty, it consists of one or more regions; when the age group configuration is not empty, it consists of one or more age groups among the preset adolescent age group, middle-aged age group, and elderly age group; the gender configuration includes male, female, and gender selection.

[0186] The model pre-training module 203 is used to pre-train the hypertrophic cardiomyopathy risk prediction model based on the first dataset.

[0187] The model fine-tuning module 204 is used to fine-tune the hypertrophic cardiomyopathy risk prediction model based on the second dataset after pre-training.

[0188] The model application module 205 is used to input the set of myocardial strain parameters of the subjects that meet the characteristics of the specified population as the corresponding myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model after the fine-tuning is completed, and to feed back the prediction probability p of this prediction to the current user. The set of myocardial strain parameters of the subjects is a set of myocardial strain parameters of the subjects that meet the characteristics of the specified population, which consists of six types of left ventricular parameters and three types of left atrial parameters.

[0189] The processing device for a hypertrophic cardiomyopathy risk prediction model provided in this embodiment of the invention can execute the method steps in the above method embodiment. Its implementation principle and technical effect are similar, and will not be repeated here.

[0190] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. Furthermore, these modules can be implemented entirely in software via processing elements; they can be fully implemented in hardware; or some modules can be implemented by processing elements calling software, while others are implemented in hardware. For example, the model building module can be a separate processing element, or it can be integrated into a chip in the above device. Alternatively, it can be stored as program code in the memory of the above device, and its functions can be called and executed by a processing element. The implementation of other modules is similar. Moreover, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0191] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more Application Specific Integrated Circuits (ASICs), one or more Digital Signal Processors (DSPs), or one or more Field Programmable Gate Arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a Central Processing Unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together as a System-on-a-Chip (SOC).

[0192] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. This computer program product includes one or more computer instructions. When these computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the foregoing method embodiments are generated. The computer described above can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The aforementioned computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the aforementioned computer instructions can be transmitted from one website, computer, server, or data center to another via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, Bluetooth, microwave, etc.) means. The aforementioned computer-readable storage medium can be any available medium that a computer can access, or a data storage device such as a server or data center that integrates one or more available media. The aforementioned available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0193] Figure 4 This is a schematic diagram of an electronic device provided in Embodiment 3 of the present invention. This electronic device can be a terminal device or server implementing the methods of the aforementioned embodiments, or it can be a terminal device or server connected to the aforementioned terminal device or server implementing the methods of the aforementioned embodiments. Figure 4 As shown, the electronic device may include: a processor 301 (e.g., CPU), a memory 302, and a transceiver 303; the transceiver 303 is coupled to the processor 301, and the processor 301 controls the transmission and reception operations of the transceiver 303. The memory 302 may store various instructions for performing various processing functions and implementing the processing steps described in the foregoing embodiments. Preferably, the electronic device involved in the embodiments of the present invention further includes: a power supply 304, a system bus 305, and a communication port 306. The system bus 305 is used to realize communication connections between components. The communication port 306 is used for communication between the electronic device and other peripherals.

[0194] exist Figure 4The system bus 305 mentioned can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This system bus can be divided into address bus, data bus, control bus, etc. For ease of representation, it is represented by only one thick line in the figure, but this does not indicate that there is only one bus or one type of bus. The communication interface is used to enable communication between the database access device and other devices (e.g., clients, read-write libraries, and read-only libraries). Memory may include Random Access Memory (RAM) and may also include non-volatile memory, such as at least one disk storage device.

[0195] The processors mentioned above can be general-purpose processors, including central processing units (CPUs), network processors (NPs), graphics processing units (GPUs), etc.; they can also be digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0196] It should be noted that the embodiments of the present invention also provide a computer-readable storage medium storing instructions that, when run on a computer, cause the computer to perform the methods and processes provided in the above embodiments.

[0197] This invention provides a method, apparatus, electronic device, and computer-readable storage medium for processing a hypertrophic cardiomyopathy (HCM) risk prediction model. As described above, this invention constructs a HCM risk prediction model capable of predicting risk based on six types of left ventricular parameters (GLS, GCS, GRS, cTSD, PSSR, and PDSR) and three types of left atrial parameters (Ee, sRe, and sRa). It also constructs a generalized population model dataset (first dataset) and a specified population model dataset (second dataset) through big data collection. The HCM risk prediction model is first pre-trained based on the first dataset and then fine-tuned based on the second dataset. After fine-tuning, the HCM risk prediction model is used to predict the risk based on the myocardial strain parameters of the subjects. This invention improves both prediction accuracy and prediction stability.

[0198] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented in hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0199] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for processing a risk prediction model for hypertrophic cardiomyopathy, characterized in that, The method includes: A deep learning model is constructed to predict the risk probability of hypertrophic cardiomyopathy, denoted as the corresponding hypertrophic cardiomyopathy risk prediction model. This model predicts the risk probability of hypertrophic cardiomyopathy based on the input myocardial strain parameter vector X and outputs the corresponding predicted probability p. The myocardial strain parameter vector X consists of six types of left ventricular parameters and three types of left atrial parameters. The six types of left ventricular parameters include GLS, GCS, GRS, cTSD, PSSR, and PDSR parameters. The three types of left atrial parameters include Ee, sRe, and sRa parameters. The predicted probability p is a probability value between 0 and 1. A generalized population model dataset is constructed as the first dataset; and a specified population model dataset is constructed as the second dataset based on preset specified population characteristics; the first dataset includes multiple first data records; the first data record includes a first training parameter vector and a first label probability; the second dataset includes multiple second data records; the second data record includes a second training parameter vector and a second label probability; both the first and second training parameter vectors are composed of the six types of left ventricular parameters and the three types of left atrial parameters; both the first and second label probabilities are label probabilities with a value of 0 or 1; the specified population characteristics include regional configuration, age group configuration, and gender configuration; when the regional configuration is not empty, it consists of one or more regions; when the age group configuration is not empty, it consists of one or more age groups selected from the preset adolescent age group, middle-aged age group, and elderly age group; the gender configuration includes male, female, and gender-neutral selection; The hypertrophic cardiomyopathy risk prediction model is pre-trained based on the first dataset; After pre-training, the hypertrophic cardiomyopathy risk prediction model is fine-tuned based on the second dataset; After fine-tuning, the set of myocardial strain parameters of the test subjects that meet the characteristics of the specified population, input by the user, is used as the corresponding myocardial strain parameter vector X and input into the hypertrophic cardiomyopathy risk prediction model for prediction. The predicted probability p output by this prediction is fed back to the current user. The set of myocardial strain parameters of the test subjects is a set of myocardial strain parameters of the test subjects that meet the characteristics of the specified population, which consists of the six types of left ventricular parameters and the three types of left atrial parameters. The hypertrophic cardiomyopathy risk prediction model has an input terminal for receiving the myocardial strain parameter vector X and an output terminal for outputting the corresponding prediction probability p. The hypertrophic cardiomyopathy risk prediction model includes a standardization module, a ventricular feature extraction module, an atrial feature extraction module, a feature splicing module, a multi-head self-attention module, a residual connection module, a feature mapping module, and a Sigmoid function module. The input of the standardization module is connected to the input of the model, and the first and second outputs are connected to the inputs of the ventricular and atrial feature extraction modules, respectively. The outputs of the ventricular and atrial feature extraction modules are connected to the first and second inputs of the feature splicing module, respectively. The output of the feature splicing module is connected to the input of the multi-head self-attention module and the first input of the residual connection module, respectively. The output of the multi-head self-attention module is connected to the second input of the residual connection module. The output of the residual connection module is connected to the input of the feature mapping module. The output of the feature mapping module is connected to the input of the Sigmoid function module. The output of the Sigmoid function module is connected to the model output.

2. The processing method of the hypertrophic cardiomyopathy risk prediction model according to claim 1, characterized in that, The standardization module is used to extract the corresponding six types of left ventricular parameters and three types of left atrial parameters from the myocardial strain parameter vector X; and to standardize the nine types of myocardial strain parameters according to preset standardization rules to obtain the corresponding six types of standardized left ventricular parameters and three types of standardized left atrial parameters; and to form the corresponding left ventricular parameter vector X from the six types of standardized left ventricular parameters. V The left atrial parameter vector X, composed of the three types of standardized left atrial parameters, is sent to the ventricular feature extraction module. A Send to the atrial feature extraction module; wherein, the standardization rules include at least the z-score rule; the six types of left ventricular standardized parameters include GLS standardized parameters, GCS standardized parameters, GRS standardized parameters, cTSD standardized parameters, PSSR standardized parameters, and PDSR standardized parameters; the three types of left atrial standardized parameters include Ee standardized parameters, sRe standardized parameters, and sRa standardized parameters; the left ventricular parameter vector X V It consists of 6 parameters Composition, 1 ≤ index i ≤ 6, 6 parameters Each of the six types of standardized left ventricular parameters corresponds one-to-one; the left atrial parameter vector X A It consists of 3 parameters Composition, 1 ≤ index j ≤ 3, 3 parameters Each of the three types of standardized parameters of the left atrium corresponds one-to-one; The ventricular feature extraction module consists of two fully connected layers connected sequentially, with a Leaky ReLU activation function following each fully connected layer; the ventricular feature extraction module is used to process the left ventricular parameter vector X. V Feature extraction is performed to obtain the corresponding feature vector H. V Send to the feature splicing module; The feature vector H V The reasoning process is as follows: , W V,1 b V,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer of the ventricular feature extraction module. V,2 b V,2 The weight matrix parameters and offset vector parameters are those of the second fully connected layer of the ventricular feature extraction module; the left ventricular parameter vector X V The shape is 6×1, and the weight matrix parameter W V,1 The shape is C V,1 ×6, the offset vector parameter b V,1 The shape is C V,1 ×1, LeakyReLU(W V,1 X V +b V,1 The shape of the output vector is C. V,1 ×1; the weight matrix parameter W V,2 The shape is C V,2 ×C V,1 The offset vector parameter b V,2 The shape is C V,2 ×1, the feature vector H V The shape is C V,2 ×1;C V,1 C V,2 These are two preset feature vector dimensions, and 6 < C. V,1 <C V,2 ; The atrial feature extraction module consists of two fully connected layers connected sequentially, with a LeakyReLU activation function following each fully connected layer; the atrial feature extraction module is used to process the left atrial parameter vector X. A Feature extraction is performed to obtain the corresponding feature vector H. A Send to the feature splicing module; The feature vector H A The reasoning process is as follows: ， W A,1 b A,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer of the atrial feature extraction module. A,2 b A,2 The weight matrix parameters and offset vector parameters are those of the second fully connected layer of the atrial feature extraction module; the left atrial parameter vector X A The shape is 3×1, and the weight matrix parameter W A,1 The shape is C A,1 ×3, the offset vector parameter b A,1 The shape is C A,1 ×1, LeakyReLU(W A,1 X A +b A,1 The shape of the output vector is C. A,1 ×1; the weight matrix parameter W A,2 The shape is C A,2 ×C A,1 The offset vector parameter b A,2 The shape is C A,2 ×1, the feature vector H A The shape is C A,2 ×1;C A,1 C A,2 These are two preset feature vector dimensions, and 3 < C. A,1 <C A,2 C A,1 <C V,1 C A,2 <C V,2 ; The feature concatenation module is used to concatenate the feature vector H. V and the feature vector H A The corresponding concatenated vector H is obtained by performing vector concatenation. C The concatenation vector H is sent to the multi-head self-attention module and the residual connection module. C The shape is C C ×1, C C For a predefined feature vector dimension, C C =C V,2 +C A,2 ; The multi-head self-attention module consists of N head It consists of N parallel single-head attention units and one multi-head fusion unit; head The preset total number of single heads, N head A feature vector dimension C is greater than 1. C Divisible positive integers; each of the single-head attention units corresponds to a set of independent query / key / value matrix parameters. / / 1 ≤ index n ≤ N head Each group of query / key / value matrix parameters / / The shape is C head ×C C C head For a predefined feature vector dimension, C head =C C / N head The multi-head fusion unit corresponds to a fusion matrix parameter W. R The fusion matrix parameter W R The shape is C C ×C C ; The multi-head self-attention module is used to process the concatenated vector H. C Perform multi-head self-attention operation to obtain the corresponding feature vector H R Sending to the residual connection module specifically involves each of the single-head attention units using the corresponding query / key / value matrix parameters. / / For the splicing vector H C Perform a self-attention operation to obtain the corresponding single-head feature vector H. n Send to the multi-head fusion unit; the multi-head fusion unit then processes N. head The single-head feature vector H n Feature concatenation is performed to obtain the corresponding concatenation vector H. all-head And based on the fusion matrix parameter W R For the splicing vector H all-head The corresponding feature vector H is obtained by performing feature mapping processing. R Send to the residual connection module; Each of the single-head feature vectors H n The reasoning process is as follows: ， , ， ； Q n K n V n Let the query vector, key vector, and value vector be the nth single-head attention unit. Let K be the key vector. n The transpose of the vector, softmax() is the Softmax function; the query, key, and value vector Q n K n V n The shape is C head ×1, the transpose vector The shape is 1×C head The single-head feature vector H n The shape is C head ×1; The splicing vector H all-head The shape is (C head ×N head )×1=C C ×1; The feature vector H R The reasoning process is as follows: ， The feature vector H R The shape is C C ×1; The residual connection module is used to process the splicing vector H according to the residual connection method. C and the feature vector H R The corresponding feature vector H is obtained by fusion. R+C Send to the feature mapping module; The feature vector H R+C The reasoning process is as follows: ， The feature vector H R+C The shape is C C ×1; The feature mapping module consists of two fully connected layers connected sequentially, with a Swish activation function following the first fully connected layer; the feature mapping module is used to map the feature vector H... R+C The confidence score is calculated and the corresponding confidence score z is sent to the Sigmoid function module; The reasoning process for the confidence score z is as follows: , W out,1 b out,1 W represents the weight matrix parameters and offset vector parameters of the first fully connected layer of the feature mapping module. out,2 b out,2 The weight matrix parameters and offset vector parameters are those of the second fully connected layer of the feature mapping module; the weight matrix parameter W out,1 The shape is C out,1 ×C C The offset vector parameter b V,1 The shape is C out,1 ×1, Swish(W out,1 H R+C +b out,1 The shape of the output vector is C. out,1 ×1; the weight matrix parameter W out,2 The shape is C out,2 ×C out,1 The offset vector parameter b out,2 The shape is C out,2 ×1, C out,1 C out,2 These are two preset feature vector dimensions, C. out,1 >C C C out,2 =1; the confidence score z is a scalar; The Sigmoid function module is used to input the confidence score z into the Sigmoid function for calculation and output the calculation result as the corresponding prediction probability p.

3. The processing method of the hypertrophic cardiomyopathy risk prediction model according to claim 1, characterized in that, The dataset used to construct the generalized population model, serving as the first dataset, specifically includes: Step 31: Collect large data of cardiac magnetic resonance images of multiple populations of multiple ages in multiple regions through preset multiple data collection channels to obtain the corresponding original image set. The data collection channels include at least publicly available medical image datasets, publicly available or authorized medical institution databases, and publicly available or authorized medical image sharing platforms; the regions include at least two or more regions; the age groups include at least adolescents, middle-aged individuals, and the elderly; the population groups include at least healthy individuals without any heart disease, other patients without hypertrophic cardiomyopathy but with other heart diseases, and a target patient population with hypertrophic cardiomyopathy; the gender of the population groups should cover both male and female; the original image set includes multiple original image records; each original image record includes a first image, a first region, a first gender, a first age group, and a first population type; the first image is a cardiac magnetic resonance imaging (MRI) image; the first region is one of the multiple regions; the first gender includes males and females; the first age group includes adolescents, middle-aged individuals, and the elderly; the first population type includes healthy individuals without any heart disease, other patients without hypertrophic cardiomyopathy but with other heart diseases, and a target patient population with hypertrophic cardiomyopathy. Step 32: Take the first image of each original image record in the original image set as the corresponding current image; and quantitatively evaluate the six types of left ventricular parameters and the three types of left atrial parameters corresponding to the current image based on a preset myocardial strain parameter evaluation software, and form a corresponding first training parameter vector from the evaluated six types of left ventricular parameters and the three types of left atrial parameters; identify whether the first population type corresponding to the current image is a target patient population with hypertrophic cardiomyopathy, and if so, set a corresponding first label probability of 1, otherwise set a corresponding first label probability of 0; and form a corresponding first data record from the first training parameter vector corresponding to the current image and the first label probability; and form a corresponding first dataset from all the obtained first data records. The myocardial strain parameter assessment software includes at least CVI42 software, CAAS MR Strain software, and Segment software.

4. The processing method of the hypertrophic cardiomyopathy risk prediction model according to claim 3, characterized in that, The construction of a specified population model dataset based on preset specified population characteristics as the corresponding second dataset specifically includes: Extract the corresponding region configuration, age group configuration, and gender configuration from the specified population characteristics; The system identifies whether the region configuration is empty; if so, the original image set is used as the corresponding first image set; otherwise, the original image records in the original image set that satisfy the region configuration in the first region are used to form the corresponding first image set. The system identifies whether the age group configuration is empty; if so, the first image set is used as the corresponding second image set; if not, the original image records in the first image set that satisfy the age group configuration are used to form the corresponding second image set. The gender configuration is identified; if the gender configuration is male, the original image records of the first gender being male in the second image set are used to form the corresponding third image set; if the gender configuration is female, the original image records of the first gender being female in the second image set are used to form the corresponding third image set; if the gender configuration is both male and female, the second image set is used as the corresponding third image set. The first data record corresponding to each original image record in the third image set is extracted from the first dataset and used as a corresponding second data record; and all the obtained second data records are used to form the corresponding second dataset.

5. The processing method of the hypertrophic cardiomyopathy risk prediction model according to claim 1, characterized in that, The pre-training of the hypertrophic cardiomyopathy risk prediction model based on the first dataset specifically includes: Step 51: Based on the preset first segmentation ratio, the first dataset is randomly segmented into a first training set and a first evaluation set; the full model parameter set of the hypertrophic cardiomyopathy risk prediction model is used as the corresponding current model parameter set; and the first training set is divided into multiple first batch record sets according to the preset maximum number of records per batch. Wherein, both the first training set and the first evaluation set consist of multiple first data records; the ratio of the total number of records in the first training set to the total number of records in the first evaluation set satisfies the first segmentation ratio; the total number of first data records in the first batch of record sets is less than or equal to the maximum number of records in a single batch; Step 52: Take the first batch of records as the corresponding current record set; Step 53: Perform a traversal of all the first data records in the current record set; during this traversal, the first data record currently traversed is used as the corresponding current training record; the first training parameter vector of the current training record is used as the current myocardial strain parameter vector X and input into the hypertrophic cardiomyopathy risk prediction model for prediction, and the prediction probability p output by this prediction is used as the corresponding first prediction probability; the current first prediction probability and the first label probability of the current training record form a corresponding first prediction-label pair; at the end of this traversal, all the first prediction-label pairs obtained in this traversal are substituted into a preset first model loss function to calculate the corresponding first loss value; The first model loss function is implemented based on the L1 loss function, the L2 loss function, or the binary cross-entropy loss function. Step 54: Identify whether the first loss value meets the preset first loss value range; if the first loss value meets the first loss value range, identify whether the current record set is the last of the first batch of record sets in the first training set; if so, proceed to step 55; otherwise, take the next batch of record sets as the new current record set and return to step 53; if the first loss value does not meet the first loss value range, perform a round of modulation on the current model parameter set based on the preset first model optimizer in the direction of minimizing the first model loss function, and return to step 53 after the end of this round of modulation. The first model optimizer includes at least the AdamW optimizer, the Adam optimizer, and the SGD optimizer. Step 55: Perform a traversal of all the first data records in the first evaluation set; during this traversal, take the currently traversed first data record as the corresponding current evaluation record; and input the first training parameter vector of the current evaluation record as the current myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and take the prediction probability p output by this prediction as the corresponding second prediction probability; and form a corresponding second prediction-label pair with the current second prediction probability and the first label probability of the current evaluation record; and at the end of this traversal, input all the obtained second prediction-label pairs into the preset first model evaluation function to calculate the corresponding first evaluation value; The first model evaluation function is implemented based on the MAE function, MSE function, or RMSE function. Step 56: Identify whether the first evaluation value meets the preset first evaluation value range; if not, return to step 52; if it does, confirm that the model pre-training is complete.

6. The processing method of the hypertrophic cardiomyopathy risk prediction model according to claim 1, characterized in that, The fine-tuning of the hypertrophic cardiomyopathy risk prediction model based on the second dataset specifically includes: Step 61: Based on a preset second segmentation ratio, the second dataset is randomly subdivided into corresponding second training sets and second evaluation sets; and the fusion matrix parameters W of the multi-head self-attention module of the hypertrophic cardiomyopathy risk prediction model are used. R and the weight matrix parameter W of the feature mapping module out,1 W out,2 The corresponding current fine-tuning parameter set is formed; and the second training set is divided into multiple second batch record sets according to the preset maximum number of records per batch; The second training set and the second evaluation set are each composed of multiple second data records; the ratio of the total number of records in the second training set to the total number of records in the second evaluation set satisfies the second partitioning ratio; the total number of second data records in the second batch of record sets is less than or equal to the maximum number of records in a single batch. Step 62: Take the first set of records from the second batch as the corresponding current set; Step 63: Perform a traversal of all the second data records in the current record set; during this traversal, the currently traversed second data record is used as the corresponding current training record; the second training parameter vector of the current training record is used as the current myocardial strain parameter vector X and input into the hypertrophic cardiomyopathy risk prediction model for prediction, and the prediction probability p output by this prediction is used as the corresponding third prediction probability; the current third prediction probability and the second label probability of the current training record form a corresponding third prediction-label pair; at the end of this traversal, all the third prediction-label pairs obtained in this traversal are substituted into the preset second model loss function to calculate the corresponding second loss value; The second model loss function is based on the binary cross-entropy loss function. Step 64: Identify whether the second loss value meets the preset second loss value range; if the second loss value meets the second loss value range, identify whether the current record set is the last second batch of record sets in the second training set; if so, proceed to step 65; otherwise, take the next second batch of record sets as the new current record set and return to step 63; if the second loss value does not meet the second loss value range, perform a round of fine-tuning on the current fine-tuning parameter set based on the preset second model optimizer in the direction of minimizing the second model loss function, and return to step 63 after the fine-tuning is completed. The second model optimizer includes at least the AdamW optimizer, the Adam optimizer, and the SGD optimizer; Step 65: Perform a round of traversal on all the second data records in the second evaluation set; during this round of traversal, take the currently traversed second data record as the corresponding current evaluation record; and input the second training parameter vector of the current evaluation record as the current myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and take the prediction probability p output by this prediction as the corresponding fourth prediction probability; and form a corresponding fourth prediction-label pair by the current fourth prediction probability and the second label probability of the current evaluation record; and at the end of this round of traversal, calculate the corresponding first accuracy, first precision, first recall, and first F1 score based on all the obtained fourth prediction-label pairs. Step 66: Identify whether the first accuracy, first precision, first recall, and first F1 score each satisfy the corresponding first accuracy range, first precision range, first recall range, and first F1 score range; if not, return to step 62; if yes, confirm that the model fine-tuning is complete.

7. An apparatus for performing a processing method for a hypertrophic cardiomyopathy risk prediction model according to any one of claims 1-6, characterized in that, The device includes: a model building module, a dataset building module, a model pre-training module, a model fine-tuning module, and a model application module; The model building module is used to construct a deep learning model for predicting the risk probability of hypertrophic cardiomyopathy, denoted as the corresponding hypertrophic cardiomyopathy risk prediction model. The hypertrophic cardiomyopathy risk prediction model predicts the risk probability of hypertrophic cardiomyopathy based on the myocardial strain parameter vector X input to the model and outputs the corresponding predicted probability p. The myocardial strain parameter vector X consists of six types of left ventricular parameters and three types of left atrial parameters. The six types of left ventricular parameters include GLS parameters, GCS parameters, GRS parameters, cTSD parameters, PSSR parameters, and PDSR parameters. The three types of left atrial parameters include Ee parameters, sRe parameters, and sRa parameters. The predicted probability p is a probability value between 0 and 1. The dataset construction module is used to construct a generalized population model dataset as the corresponding first dataset; and to construct a specified population model dataset as the corresponding second dataset based on preset specified population characteristics; the first dataset includes multiple first data records; the first data record includes a first training parameter vector and a first label probability; the second dataset includes multiple second data records; the second data record includes a second training parameter vector and a second label probability; the first and second training parameter vectors are both composed of the six types of left ventricular parameters and the three types of left atrial parameters; the first and second label probabilities are both label probabilities with a value of 0 or 1; the specified population characteristics include regional configuration, age group configuration, and gender configuration; when the regional configuration is not empty, it consists of one or more regions; when the age group configuration is not empty, it consists of one or more age groups among the preset adolescent age group, middle-aged age group, and elderly age group; the gender configuration includes male, female, and gender-neutral selection; The model pre-training module is used to pre-train the hypertrophic cardiomyopathy risk prediction model based on the first dataset; The model fine-tuning module is used to fine-tune the hypertrophic cardiomyopathy risk prediction model based on the second dataset after pre-training. The model application module is used to, after fine-tuning, input the set of myocardial strain parameters of the subjects that meet the characteristics of the specified population as the corresponding myocardial strain parameter vector X into the hypertrophic cardiomyopathy risk prediction model for prediction, and feed back the prediction probability p output by this prediction to the current user; the set of myocardial strain parameters of the subjects is a set of myocardial strain parameters of subjects that meet the characteristics of the specified population, which consists of the six types of left ventricular parameters and the three types of left atrial parameters; The hypertrophic cardiomyopathy risk prediction model has an input terminal for receiving the myocardial strain parameter vector X and an output terminal for outputting the corresponding prediction probability p. The hypertrophic cardiomyopathy risk prediction model includes a standardization module, a ventricular feature extraction module, an atrial feature extraction module, a feature splicing module, a multi-head self-attention module, a residual connection module, a feature mapping module, and a Sigmoid function module. The input of the standardization module is connected to the input of the model, and the first and second outputs are connected to the inputs of the ventricular and atrial feature extraction modules, respectively. The outputs of the ventricular and atrial feature extraction modules are connected to the first and second inputs of the feature splicing module, respectively. The output of the feature splicing module is connected to the input of the multi-head self-attention module and the first input of the residual connection module, respectively. The output of the multi-head self-attention module is connected to the second input of the residual connection module. The output of the residual connection module is connected to the input of the feature mapping module. The output of the feature mapping module is connected to the input of the Sigmoid function module. The output of the Sigmoid function module is connected to the model output.

8. An electronic device, characterized in that, include: Memory, processor, and transceiver; The processor is configured to be coupled to the memory, read and execute instructions in the memory to implement the method according to any one of claims 1-6; The transceiver is coupled to the processor, and the processor controls the transceiver to send and receive messages.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions that, when executed by a computer, cause the computer to perform the method according to any one of claims 1-6.

Images