基于深度迁移学习和机器学习的ECG图像分类

ECG Image Classification Based on Deep Transfer Learning and Machine Learning

摘要: 心肌梗死是威胁人类生命健康的心血管疾病之一，为了进一步提升心肌梗死的诊断效果，本文提出一种基于深度迁移学习与传统机器学习融合的心电图图像分类方法。具体来讲，将4种主流卷积神经网络的分类性能进行对比，并筛选出最优特征提取模型，再结合传统机器学习分类器构建混合分类模型。实验结果显示，基于DenseNet121深度特征的SVM混合模型展现出最佳性能，测试准确率为86.43%，AUC为0.9762，该模型在测试集中对心肌梗死(MI)类别的识别未出现误判，展现出了极高的区分能力。本方法可有效提升心电图像分类精度，为心肌梗死的智能辅助诊断提供了可靠的技术方案。

Abstract: Myocardial infarction is one of the cardiovascular diseases threatening human life and health. To further improve the diagnostic efficacy of myocardial infarction, this paper proposes an electrocardiogram image classification method based on the fusion of deep transfer learning and traditional machine learning. Specifically, the classification performance of four mainstream convolutional neural networks is compared to select the optimal feature extraction model, which is then combined with traditional machine learning classifiers to construct a hybrid classification model. Experimental results show that the hybrid model combining DenseNet121-based deep features with SVM achieves the best performance, with a test accuracy of 86.43% and an AUC of 0.9762. This model achieves perfect discrimination for the myocardial infarction (MI) category on the test set, demonstrating a high level of discriminative ability. The proposed method can effectively improve the accuracy of ECG image classification and provide a reliable technical solution for intelligent assisted diagnosis of myocardial infarction.