多模态公开疾病数据集 结合多模态特征和机器学习方法预测心血管疾病

心电图 (ECG) 和心音图 (PCG) 在心血管疾病的早期预防和诊断中起着重要作用。随着机器学习技术的发展，从心电图和 PCG 检测心血管疾病已引起广泛关注。然而，目前可用的方法大多基于单一数据资源。开发有效的多模态机器学习方法来预测和诊断心血管疾病是可取的。在这项研究中，我们提出了一种基于 ECG 和 PCG 特征预测心血管疾病的新型多模态方法。通过构建传统的神经网络，我们分别提取 ECG 和 PCG 深度编码特征。遗传算法用于筛选组合特征并获得最佳特征子集。然后支持向量机做出分类决策。实验结果表明，与使用单模态特征ECG和PCG相比，该方法在使用多模态数据资源时的性能达到了0.936的AUC值。 该数据集是由 PhysioNet/CinC Challenge 在 2016 年组装的真实数据集开发的。原始数据集可以从网站 (http://www.physionet.org/challenge/2016/) 下载。

Electrocardiography (ECG) and phonocardiography (PCG) play critical roles in the early prevention and diagnosis of cardiovascular diseases. With the advancement of machine learning technologies, detecting cardiovascular diseases using ECG and PCG signals has garnered widespread research attention. However, most existing methods are based on a single data modality. Developing effective multimodal machine learning approaches for cardiovascular disease prediction and diagnosis is highly desirable. In this study, we propose a novel multimodal method for cardiovascular disease prediction based on integrated ECG and PCG features. By constructing conventional neural networks, we separately extract deep encoded features from ECG and PCG signals. Genetic algorithms are employed to screen the fused features and obtain the optimal feature subset. Subsequently, a support vector machine (SVM) is used to make classification decisions. Experimental results demonstrate that compared with single-modality methods using only ECG or PCG features, the proposed method achieves an AUC value of 0.936 when leveraging multimodal data resources. This dataset is developed from the real-world dataset assembled by the PhysioNet/CinC Challenge in 2016. The original dataset can be downloaded from the official website (http://www.physionet.org/challenge/2016/).

搜集汇总

背景与挑战

背景概述

该数据集是一个多模态公开疾病数据集，专注于结合心电图（ECG）和心音图（PCG）的特征，利用机器学习方法预测心血管疾病。数据集大小为701M，来源于PhysioNet/CinC Challenge 2016，实验结果显示其多模态方法的AUC值达到0.936。