Hangaragi Shivalila, Neelima N, Jegdic Katarina, Nagarwal Amitesh
Department of Electronics and Communication, Amrita School of Engineering-Bangalore, Amrita Vishwa Vidyapeetham, Bangalore, India.
Department of Mathematics and Statistics, University of Houston, One Main Street, Houston, Downtown, 77002, TX, USA.
Sci Rep. 2025 Mar 8;15(1):8129. doi: 10.1038/s41598-025-92395-w.
Detection and classification of cardiovascular diseases are crucial for early diagnosis and prediction of heart-related conditions. Existing methods rely on either electrocardiogram or phonocardiogram signals, resulting in higher false positive rates. Solely ECG misses the murmurs associated with the narrowing of the blood vessels caused by abnormalities in the heart. Similarly, considering only PCG will miss the subtle changes in the electrical activity of the heart that leads to incomplete evaluation. The implementation of a multi-class heart disease classification model utilizing both ECG and PCG signals is the objective of the proposed study. The approach involves preprocessing, fusion, waveform detection utilizing the Pan-Tompkins Algorithm, and signal localization using Algebraic Integer-quantized Stationary Wavelet Transform. Low-rank Kernelized Density-Based Spatial Clustering of Applications with noise is employed to cluster signals into normal and abnormal categories. Feature selection is performed with Heming Wayed Polar Bear Optimization, and classification is done using C squared Pool Sign BI-power-activated Deep Convolutional Neural Network. The proposed model achieves a classification accuracy of 97% with 0.03 error rate. The multi-class classifier effectively identifies and classifies the heart diseases into Aortic stenosis Valvular disorder, Tricuspid Valvular disorder, Mitralstenosis Valvular disorder, Pulmonary Valvular disorder, Atrial Fibrillation, and Ischemic heart disorder.
心血管疾病的检测和分类对于心脏相关疾病的早期诊断和预测至关重要。现有方法要么依赖心电图信号，要么依赖心音图信号，导致假阳性率较高。仅依靠心电图会遗漏与心脏异常引起的血管狭窄相关的杂音。同样，仅考虑心音图会遗漏心脏电活动中的细微变化，从而导致评估不完整。本研究的目标是实现一种利用心电图和心音图信号的多类心脏病分类模型。该方法包括预处理、融合、使用Pan-Tompkins算法进行波形检测以及使用代数整数量化平稳小波变换进行信号定位。采用带噪声的基于低秩核密度的空间聚类应用将信号聚类为正常和异常类别。使用赫明·韦德北极熊优化算法进行特征选择，并使用C平方池符号双功率激活深度卷积神经网络进行分类。所提出的模型实现了97%的分类准确率，错误率为0.03。该多类分类器能够有效地将心脏病识别并分类为主动脉瓣狭窄、瓣膜疾病、三尖瓣瓣膜疾病、二尖瓣狭窄瓣膜疾病、肺动脉瓣瓣膜疾病、心房颤动和缺血性心脏病。
