A classification method for heart failure with left bundle branch block based on feature fusion

Abstract

This invention discloses a classification method for heart failure complicated with left bundle branch block based on feature fusion. The method utilizes cardiac magnetic resonance images to extract radiomics features and deep learning features, performs feature fusion and selection, and constructs a classification model using machine learning algorithms. The selected fused features are used as input. This method combines the advantages of deep learning in image processing with the efficiency of machine learning in classification tasks. By combining radiomics features and deep learning features, it can more comprehensively and accurately reflect the complex information of cardiac images. Furthermore, by using automated and intelligent feature extraction and classification methods, it can quickly process and analyze large amounts of medical image data, achieving accurate classification of heart failure complicated with left bundle branch block. This invention has higher classification accuracy and stronger robustness, contributing to improved efficiency and quality of medical services in the field of medical image classification.

Description

Technical Field

[0001] This invention relates to the field of medical image classification technology, specifically to a classification method for heart failure combined with left bundle branch block based on feature fusion. Background Technology

[0002] Heart failure (HF) is the end stage of various cardiovascular diseases, characterized by high incidence, high hospitalization rates, poor quality of life for patients, and high medical costs. Left bundle branch block (LBBB) is relatively common in HF patients. The accurate classification of LBBB and non-LBBB plays a crucial role in decision-making and prognosis. Currently, LBBB is mainly classified clinically through electrocardiograms and imaging examinations. However, this method has problems such as strong subjectivity and insufficient accuracy, which may lead to inconsistent classification results. In order to improve the accuracy of LBBB classification, cardiovascular imaging technology has received widespread attention, especially cardiac magnetic resonance imaging (CMR), which can assess myocardial deformation in detail and provide new possibilities for the accurate classification of LBBB.

[0003] Machine learning and deep learning are developing rapidly in the medical field, and medical image classification is one of its important applications. Radiomics combines medical imaging with machine learning, using computer algorithms and image processing techniques to extract a large number of quantitative features from medical images, including morphological features, texture features, density features, edge features, etc. These features can be used to classify different diseases, predict disease progression, and evaluate treatment response. As a branch of machine learning, deep learning can automatically learn and extract high-level features from data by building deep neural network models, and is particularly suitable for processing complex medical image data. Therefore, how to combine CMR with machine learning and deep learning to accurately classify patients with HF combined with LBBB remains an urgent problem to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this invention provides a feature fusion-based classification method for heart failure combined with left bundle branch block (LBBB). This method combines radiomics features from CMR images with deep learning features, and through machine learning algorithms, it can accurately classify heart failure combined with LBBB, providing a reliable basis for clinical diagnosis and treatment decisions. Specifically, it includes the following steps:

[0005] Step a: Construct a dataset, which includes cardiac magnetic resonance images of heart failure combined with left bundle branch block, and preprocess the medical image data;

[0006] Step b: Extract a large number of radiomics features from the preprocessed image, including morphological features, texture features, density features, and edge features;

[0007] Step c: After fine-tuning the pre-trained deep learning model, extract features from the pre-processed image to obtain deep learning features;

[0008] Step d: Combine radiomics features and deep learning features to perform feature fusion, obtain the fused feature vector, and then perform feature selection;

[0009] Step e: Model the selected features, evaluate and train the model to verify the performance of the fused feature model, and output the model evaluation results.

[0010] Step f: Use the model evaluation results to classify the input heart failure combined with left bundle branch block image, and obtain and output the classification results.

[0011] Preferably, in step a, the image preprocessing includes denoising, contrast enhancement, extraction of myocardial contour, and generation of a mask image of the region of interest, with the image size standardized to 256*256.

[0012] Preferably, in step b, radiomics feature extraction is implemented using the PyRadiomics open-source library. These features include gradient features, shape features, texture features, intensity features, and wavelet features, as well as image derivation processing such as gradient, exponential, logarithmic, square, square root, and wavelet filtering.

[0013] Preferably, step c, extracting deep learning features, specifically includes: using a ResNet50 model pre-trained on the ImageNet dataset, fine-tuning the ResNet50 model on the ACDC2017 heart dataset to optimize the model's classification ability for heart image features; using the fine-tuned ResNet50 model to extract features from the input heart image; selecting the output of the penultimate average pooling layer as the image feature; the average pooling layer is responsible for adjusting the output of the previous layer into a fixed-size average pooling feature map; its output features can capture key morphological, texture, and spatial relationship information in the heart image; and using the extracted features in subsequent analysis and classification tasks to achieve accurate diagnosis of heart diseases.

[0014] Preferably, in step d, feature selection is achieved through a combination of statistical analysis and machine learning techniques, including: using t-test to screen the initial fused features, selecting features with p-values ​​less than 0.05 as important features, then applying Pearson correlation coefficient analysis to further evaluate the correlation between these features, removing redundant features with correlation coefficients greater than 0.9 and retaining the most representative feature subset, and then applying the LASSO model to compress and select the remaining features. This model, by applying an L1 regularization term, can automatically select the features that contribute the most to the classification task while ensuring model performance, thus obtaining the final non-zero coefficient feature set.

[0015] Preferably, in step e, the established model includes LR, Bayes, SVM, and KNN use stratified sampling to divide the dataset into training and test sets in a 7:3 ratio. The training set is used to build and train the prediction model, while the test set is used to evaluate the model's performance and generalization ability.

[0016] Preferably, in the model evaluation, after training is completed, the model is evaluated using a test set. The evaluation metrics include ROC curve and AUC value, accuracy, sensitivity, specificity and F1 score. Decision curve analysis (DCA) is also used to evaluate the net benefit of the model.

[0017] Preferably, in the verification that the fusion feature model is superior to the radiomics feature model and the deep learning model, the radiomics feature model and the deep learning model are modeled and evaluated separately, and compared with the fusion feature model to verify whether they can enrich the feature representation and whether they can enable the model to capture more diverse information.

[0018] Preferably, in step f, the method of classifying the input heart failure combined with left bundle branch block image using the model evaluation results specifically includes: inputting the heart failure combined with left bundle branch block cardiac magnetic resonance image into the trained fusion feature model, using the fusion feature model to classify and predict the heart failure combined with left bundle branch block cardiac magnetic resonance image, and outputting the classification result of the classification prediction.

[0019] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0020] 1. By combining radiomics features and deep learning features, the complex information of cardiac images can be reflected more comprehensively and accurately, improving the classification accuracy of heart failure complicated with left bundle branch block;

[0021] 2. By utilizing automated and intelligent feature extraction and classification methods, large amounts of medical image data can be processed and analyzed quickly, improving diagnostic efficiency;

[0022] 3. To provide clinicians with objective and accurate diagnostic information, assist them in developing personalized treatment plans, and improve treatment outcomes and patients' quality of life. Attached Figure Description

[0023] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.

[0024] In the attached diagram:

[0025] Figure 1 This is a schematic diagram of the overall process provided for an embodiment of the present invention;

[0026] Figure 2 This is a schematic diagram of dataset partitioning provided in an embodiment of the present invention;

[0027] Figure 3 This is a schematic diagram of feature extraction provided in an embodiment of the present invention;

[0028] Figure 4 This is a schematic diagram of the feature selection results provided in an embodiment of the present invention;

[0029] Figure 5 A schematic diagram comparing the ROC curves of various models provided in the embodiments of the present invention;

[0030] Figure 6 A schematic diagram comparing the DCA curves of various models provided in the embodiments of the present invention;

[0031] Figure 7 This is a schematic diagram of the confusion matrix classification results of the fusion feature model provided in an embodiment of the present invention. Detailed Implementation

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention without making any creative efforts shall fall within the scope of protection of the present invention.

[0033] This invention combines Figures 1 to 7 The following technical solution is provided: a classification method for heart failure combined with left bundle branch block based on feature fusion, the specific operation process of which includes:

[0034] Step a: Construct a dataset including cardiac magnetic resonance images of heart failure combined with left bundle branch block, and preprocess the medical image data in the dataset;

[0035] In this embodiment, image preprocessing includes denoising, contrast enhancement, extraction of myocardial contours, and generation of mask images for regions of interest. The image size is standardized to 256*256. First, region of interest segmentation is performed to extract regions of interest representing the myocardium from the original image. In this embodiment, this is combined with... Figure 1 , Figure 1 The processing flow of the present invention is illustrated. Figure 1 In the image, two different images represent the original image and the mask image, respectively. The accuracy of ROI segmentation is crucial for subsequent feature extraction and model construction.

[0036] Step b: Extract radiomics features from the preprocessed image, including morphological features, texture features, density features, and edge features;

[0037] In this embodiment, after the region of interest is segmented, the next step is to extract radiomics features and deep learning features. For radiomics features, these include shape, texture, and first-order statistical features. That is, radiomics feature extraction is implemented through the PyRadiomics open-source library. These features include gradient features, shape features, texture features, intensity features, and wavelet features, as well as image derivation processing such as gradient, exponential, logarithmic, square, square root, and wavelet filtering.

[0038] Step c: Use a pre-trained deep learning model to extract features from the pre-processed image after fine-tuning to obtain deep learning features;

[0039] In this embodiment, for deep learning features, a ResNet50 model pre-trained on the ImageNet dataset is selected, and after fine-tuning on the ACDC2017 heart dataset, the output of the penultimate average pooling layer is selected as the image features. These features can reflect the shape, size, texture, spatial relationship and other attributes of the region of interest.

[0040] For example, extracting deep learning features specifically includes: using a ResNet50 model pre-trained on the ImageNet dataset, fine-tuning the ResNet50 model on the ACDC2017 heart dataset to optimize the model's ability to classify heart image features, using the fine-tuned ResNet50 model to extract features from the input heart image, selecting the output of the penultimate average pooling layer as the image feature, the average pooling layer being responsible for adjusting the output of the previous layer into a fixed-size average pooling feature map, whose output features can capture key morphological, texture, and spatial relationship information in the heart image, and using the extracted features in subsequent analysis and classification tasks to achieve accurate diagnosis of heart diseases.

[0041] Step d: Combine radiomics features and deep learning features to perform feature fusion, obtain the fused feature vector, and then perform feature selection;

[0042] In this embodiment, after fusing the two types of features, the next step is feature selection. The purpose of feature selection is to select the most representative features that best reflect disease information from a large number of features, so as to reduce the complexity of the model and improve the generalization ability of the model. Here, t-test, Pearson and LASSO algorithms are used for feature selection. t-test will initially screen features based on the p-value of the features. Pearson correlation coefficient can be used to evaluate the linear correlation between features and target variables. LASSO is a linear regression method that can achieve feature selection and sparsity of weights by introducing L1 regularization term.

[0043] For example, feature selection is achieved through a combination of statistical analysis and machine learning techniques, including: using t-test to screen the initial fused features, selecting features with p-values ​​less than 0.05 as important features, then applying Pearson correlation coefficient analysis to further evaluate the correlation between these features, removing redundant features with correlation coefficients greater than 0.9 and retaining the most representative feature subset, and then applying the LASSO model to compress and select the remaining features. This model, by applying an L1 regularization term, can automatically select the features that contribute the most to the classification task while ensuring model performance, thus obtaining the final non-zero coefficient feature set.

[0044] Step e: Model the selected features, evaluate and train the model to verify the performance of the fused feature model, and output the model evaluation results.

[0045] In this embodiment, the final step is to construct the prediction model and evaluate it using test set data, constructing LR and LR models respectively. Bayes, SVM, and KNN models were used to partition the dataset using stratified sampling, dividing it into training and test sets in a 7:3 ratio. The training set was used to build and train the prediction model, while the test set was used to evaluate the model's performance and generalization ability. Model evaluation was conducted after training, using the test set to assess the model. Evaluation metrics included ROC curve and AUC value, accuracy, sensitivity, specificity, and F1 score. Decision curve analysis (DCA) was also used to evaluate the model's net gain, selecting the best-performing model. Finally, the fusion feature model was validated, demonstrating its superiority over the radiomics feature model and deep learning model. Radiomics features and deep learning models were modeled and evaluated separately, then compared with the fusion feature model to verify whether they could enrich feature representation and enable the model to capture more diverse information.

[0046] In this embodiment, combined with Figure 2 , Figure 2 The dataset is shown in the diagram. The dataset used in this embodiment comes from HF merged LBBB data instances from 3 centers, with a total of 200 HF instances, of which 73 instances exhibit LBBB mode and 127 instances exhibit non-LBBB mode.

[0047] In this embodiment, combined with Figure 3 , Figure 3 The diagram illustrates feature extraction. By constructing radiomics features, deep learning features, and fusion features separately, and then comparing them, we can verify that the fusion features have the best model prediction ability.

[0048] For example, manually extracted radiomics features can be divided into three categories: geometry, intensity, and texture. Geometric features are signal-independent features used to describe the two-dimensional and three-dimensional shape of the myocardium. Intensity features depict the first-order statistical distribution of voxel intensity in the myocardium. Texture features describe patterns, i.e., the second-order and higher-order spatial distribution of intensity. Texture features are derived using several different methods, including gray-level co-occurrence matrix (GLCM), gray-level run-length matrix (GLRLM), gray-level size region matrix (GLSZM), and neighborhood gray-level difference matrix (NGTDM).

[0049] Preferably, the CMR image is first derived using gradient, exponential, logarithmic, square, square root, and wavelet filtering (HHH, HHL, HLH, HLL, LHH, LHL, LLH, LLL), and then 1288 handcrafted features are extracted, including 306 first-order features, 14 shape features, and 968 texture features. All extracted features are normalized using the Z-Score method.

[0050] Furthermore, deep learning features were extracted using a ResNet50 network pre-trained on the ImageNet dataset. To improve the accuracy and specificity of feature extraction, the ResNet50 was fine-tuned using the ACDC2017 heart dataset, which contains CMR images of 150 patients, with several CMR images for each patient. After filtering out all images containing myocardial tissue, image preprocessing operations such as resizing and normalization were performed, resulting in a total of 2817 images with a size of 256*256. These images were then input into the ResNet50 for training for 200 epochs with a batch size of 32. After training, deep learning feature extraction was performed, and the corresponding deep learning results were obtained. Finally, the two classes were fused and feature selection was performed to select the most representative features for subsequent classification tasks.

[0051] Preferably, since many initially extracted features may be redundant or meaningless, the first step uses a t-test to reduce feature dimensionality, with a significance level of p = 0.05. If the p-value of a feature is less than 0.05, it is considered to have a significant difference and is therefore retained as an important feature for further analysis; conversely, if the p-value is greater than 0.05, the feature is deleted. The next feature selection step is Pearson correlation coefficient with a threshold of 0.9. If the correlation coefficient of two features is greater than the threshold, one is retained and the other is discarded. Finally, based on the regularization weight λ, the LASSO regression method compresses all regression coefficients to zero and assigns zero coefficients to many unrelated features. The optimal λ is found through 10-fold cross-validation. Features with non-zero coefficients obtained in this process are selected for fitting the classification model.

[0052] In this embodiment, combined with Figure 4 , Figure 4 The diagram shows the feature selection results. After selection, the fused features yielded 11 features, including 2 radiomics features and 9 deep learning features. Each feature corresponds to a coefficient value, the magnitude of which and its sign reflect the importance of the feature in the model.

[0053] Preferably, four classification models, namely LR, Naive Bayes, KNN and SVM, were constructed and trained respectively. Then, these models were evaluated using a test dataset. The evaluation metrics included AUC, accuracy, sensitivity, specificity, F1 score and decision curve analysis (DCA) to verify the predictive ability and reliability of the models. The comparison results of some metrics are shown in Table 1. It can be seen that the SVM model has the highest accuracy, specificity and F1 score on the test set, which are 0.883, 0.900 and 0.829 respectively, and the highest AUC value of 0.949.

[0054] Table 1: The performance of the four models on the training and test sets is shown in the table below;

[0055]

[0056] In this embodiment, combined with Figure 5 , Figure 5 The diagram shows a comparison of the ROC curves of various models, where A represents the training set results and B represents the test set results. The SVM model has the largest area under the curve enclosed by the training set and is also the best on the test set.

[0057] In this embodiment, combined with Figure 6 , Figure 6 This diagram illustrates the comparison of DCA curves for each model, where A, B, C, and D represent LR, ... Bayes, SVM, and KNN models use decision curves that consider patient risk and benefit to assess the feasibility of clinical decisions. The horizontal axis represents the risk threshold, and the vertical axis represents the net benefit. The larger the area enclosed by the model and the treat all and treat none curves, the higher the clinical benefit of the model. As can be seen from the figure, the SVM model has the largest enclosed area and the highest net benefit.

[0058] For example, after evaluation, SVM outperformed other models among four different machine learning models, and these results indicate that SVM has better classification capabilities.

[0059] Specifically, the comparison results of the three feature models are shown in Table 2. The radiomics feature model, modeled using an SVM model, shows a significant performance drop on the test set and relatively low overall performance, making it not the optimal choice. The deep learning model exhibits high performance on both the training and test sets, with a relatively smaller performance drop on the test set, indicating good generalization ability. However, the fusion feature model, combining the advantages of multiple features, has the highest AUC value (0.996) on the training set, 0.046 higher than the deep learning model, and maintains high performance on the test set (AUC = 0.949), demonstrating strong comprehensive predictive ability. By integrating information from radiomics and deep learning features, the fusion feature model is better at capturing the complexity and subtle differences in data, thus exhibiting superior performance.

[0060] Table 2: The comparison results of the three feature models are shown in the table below;

[0061]

[0062] Step f: Use the model evaluation results to classify the input heart failure combined with left bundle branch block image, and obtain and output the classification results.

[0063] In this embodiment, combined with Figure 7 , Figure 7 This diagram illustrates the confusion matrix classification results of the fusion feature model. A cardiac MRI image of heart failure combined with left bundle branch block is input into the trained fusion feature model. The model is then used to classify and predict the classification of this image, outputting the classification results. Compared with other methods, the model correctly identifies 3 errors in LBBB examples (sensitivity 0.959) and 5 errors in non-LBBB examples (specificity 0.961), with an overall accuracy of 0.960. This demonstrates that the fusion feature model effectively distinguishes between these two categories, enabling rapid processing and analysis of large amounts of medical image data and improving diagnostic efficiency.

[0064] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0065] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A classification method for heart failure complicated with left bundle branch block based on feature fusion, characterized in that: Includes the following steps: Step a: Construct a dataset including cardiac magnetic resonance images of heart failure combined with left bundle branch block, and preprocess the medical image data in the dataset; Step b: Extract radiomics features from the preprocessed image, including morphological features, texture features, density features, and edge features; Step c: After fine-tuning the pre-trained deep learning model, extract features from the pre-processed image to obtain deep learning features; Step d: Combine the radiomics features and the deep learning features to perform feature fusion, obtain the fused feature vector, and then perform feature selection; The feature selection process includes: using t-test to screen the initial fusion features, selecting features with p-values ​​less than 0.05 as important features, applying Pearson correlation coefficient analysis to evaluate the correlation between features, removing redundant features with correlation coefficients greater than 0.9, and then applying the LASSO model to compress and select the remaining features to obtain the final non-zero coefficient feature set. Step e: Model the selected features, evaluate and train the model to verify the performance of the fused feature model, and output the model evaluation results. Step f: Use the model evaluation results to classify the input heart failure combined with left bundle branch block image, and obtain and output the classification results.

2. The method for classifying heart failure with left bundle branch block based on feature fusion according to claim 1, characterized in that: The preprocessing procedure for the medical image data in the dataset includes: denoising, contrast enhancement, extraction of myocardial contour, generation of mask images of regions of interest, with the image size standardized to 256*256, and region of interest segmentation to extract regions of interest from the original image, representing the myocardium.

3. The method for classifying heart failure with left bundle branch block based on feature fusion according to claim 2, characterized in that: The extraction of radiomics features also includes: After the region of interest is segmented, radiomics features and deep learning features are extracted. Radiomics features include shape, texture and first-order statistical features. Radiomics feature extraction is implemented using the PyRadiomics open source library. The radiomics features include gradient features, shape features, texture features, intensity features, and wavelet features, as well as gradient, exponential, logarithmic, square, square root, and wavelet filtered image derivation processing.

4. The method for classifying heart failure with left bundle branch block based on feature fusion according to claim 3, characterized in that: The method of using a pre-trained deep learning model for feature extraction from pre-processed images after fine-tuning also includes: using a ResNet50 model pre-trained on the ImageNet dataset, fine-tuning the ResNet50 model on the ACDC2017 heart dataset to optimize the model's ability to classify heart image features. Using a finely tuned ResNet50 model, features are extracted from the input heart image, and the output of the penultimate average pooling layer is selected as the image features. The average pooling layer is responsible for adjusting the output of the previous layer to a fixed-size average pooling feature map, and its output features can capture key morphological, texture and spatial relationship information in heart images. The extracted features are used in subsequent analysis and classification tasks to achieve the diagnosis of heart diseases.

5. The method for classifying heart failure with left bundle branch block based on feature fusion according to claim 4, characterized in that: The image features are used to reflect the shape, size, texture, and spatial relationships of the region of interest.

6. The method for classifying heart failure with left bundle branch block based on feature fusion according to claim 5, characterized in that: The feature selection process includes: using t-test, Pearson correlation coefficient and LASSO algorithm to select features. The t-test initially filters features based on their p-values. The Pearson correlation coefficient is used to evaluate the linear correlation between features and the target variable. The LASSO algorithm achieves feature selection and weight sparsity by introducing an L1 regularization term.

7. The method for classifying heart failure with left bundle branch block based on feature fusion according to claim 6, characterized in that: Step e specifically includes: constructing a prediction model, evaluating the model using test set data, constructing LR, Naïve Bayes, SVM and KNN models respectively, dividing the dataset using stratified sampling, dividing it into training set and test set in a 7:3 ratio, the training set being used to construct and train the prediction model, and the test set being used to evaluate the model's performance and generalization ability. The model evaluation process includes: after training, evaluating the model using a test set, with evaluation metrics including ROC curve and AUC value, accuracy, sensitivity, specificity, and F1 score; evaluating the net benefit of the model using decision curve analysis; selecting the best-performing model; and finally validating the fusion feature model to verify its superiority over the radiomics feature model and deep learning model. This involves separately modeling and evaluating the radiomics feature and deep learning models, comparing them with the fusion feature model, and verifying whether they can enrich feature representation and enable the model to capture more diverse information.

8. The method for classifying heart failure with left bundle branch block based on feature fusion according to claim 7, characterized in that: The method for classifying the input heart failure combined with left bundle branch block image using the model evaluation results specifically includes: inputting the heart magnetic resonance image of heart failure combined with left bundle branch block into the trained fusion feature model, using the fusion feature model to classify and predict the heart magnetic resonance image of heart failure combined with left bundle branch block, and outputting the classification result of the classification prediction.

Images