Class distribution of dataset.

BackgroundHeart muscle damage from myocardial infarction (MI) is brought on by insufficient blood flow. The leading cause of death for middle-aged and older people worldwide is myocardial infarction (MI), which is difficult to diagnose because it has no symptoms. Clinicians must evaluate electrocardiography (ECG) signals to diagnose MI, which is difficult and prone to observer bias. To be effective in actual practice, an automated, and computerized detection system for Myocardial Infarction using ECG images, must meet a number of criteria.ObjectiveIn an actual clinical situation, these requirements—such as dependability, simplicity, and superior decision-making abilities—remain crucial. In the current work, we have developed a model using a dataset that consists of a combination of 928 ECG images taken from publicly available Mendeley Data. It was converted into three classes Myocardial Infarction, Abnormal heartbeat, and Normal.MethodsThe dataset is then imported, pre-processed, and split into a 70:20:10 ratio of training, validation, and testing. It is then trained using the Siamese Network Model.ResultsThe classification accuracy comes out to be 98%. The algorithm works excellently with datasets having class imbalance by taking pair of images as input. The validation and testing classification matrix is then generated and the evaluation metrics for both of them come out to be a near-perfect value.ConclusionIn this study, we developed the ECG signals based early detection of cardiovascular diseases with Siamese network model.

背景：心肌梗死（myocardial infarction, MI）引发的心肌损伤，源于血液灌注不足。心肌梗死是全球中老年人群的首要致死病因，且该疾病常无明显临床症状，诊断难度较高。临床医师需通过分析心电图（electrocardiography, ECG）信号以诊断心肌梗死，但该操作繁琐且易受观察者偏倚影响。若要在实际临床场景中切实可用，基于心电图图像的心肌梗死自动化计算机辅助检测系统需满足多项严苛标准。
目标：在实际临床场景中，系统的可靠性、易用性与优异的决策性能仍是核心要求。本研究依托公开数据集平台Mendeley Data中的928张心电图图像构建检测模型，该数据集被划分为心肌梗死（Myocardial Infarction）、异常心律（Abnormal heartbeat）与正常（Normal）三类。
方法：本研究首先导入数据集并完成预处理，随后按照70:20:10的比例将数据集划分为训练集、验证集与测试集，并基于孪生网络（Siamese Network）模型开展模型训练。
结果：本模型的分类准确率达98%。由于该算法以图像对作为输入，其在类别不平衡数据集上亦表现优异。随后本研究生成了验证集与测试集的分类混淆矩阵，两类数据集的各项评估指标均接近最优值。
结论：本研究基于孪生网络模型，构建了基于心电图信号的心血管疾病早期检测方案。