Heart sound signal classification method based on spectral envelope feature extraction(PDF)
《中国医学物理学杂志》[ISSN:1005-202X/CN:44-1351/R]
- Issue:
- 2026年第7期
- Page:
- 958-964
- Research Field:
- 医学信号处理与医学仪器
- Publishing date:
Info
- Title:
- Heart sound signal classification method based on spectral envelope feature extraction
- Author(s):
- YI Minsheng1; 2; LENG Hongxia2; LIU Xing2; FANG Yu2
- 1. Shaoyang Polytechnic, Shaoyang 422000, China 2. School of Electrical and Electronic Information, Xihua University, Chengdu 610039, China
- Keywords:
- Keywords: heart sound classification power spectrum intensity feature extraction heart valve disease frequency domain envelope
- PACS:
- 心音分类；功率谱强度；特征提取；心脏瓣膜疾病；频域包络
- DOI:
- DOI:10.3969/j.issn.1005-202X.2026.07.017
- Abstract:
- Abstract: Cardiovascular diseases can predispose the heart to structural and functional abnormalities, leading to weakened myocardial contractility and reduced blood-pumping capacity, which severely jeopardizes patients physical and psychological health. Therefore, this study proposes a heart sound analysis method based on frequency-balanced power spectrum intensity (FBPSI) features to enable effective differentiation of normal and pathological heart sound signals. An adaptive wavelet threshold shrinkage denoising algorithm is first employed to preprocess heart sound signals and extract FBPSI envelopes. Subsequently, multi-dimensional features are extracted from the FBPSI envelopes, which are then fused with features derived from power spectral density and energy spectral density. The importance of these features is ranked using the minimum redundancy maximum relevance algorithm. Finally, machine learning models are applied for the classification and identification of heart sound signals. Experimental validation is conducted using the Yaseen and the PhysioNet/CinC Challenge 2016 public datasets. Ten-fold cross-validation results demonstrate that the proposed effective feature vectors combined with the K-nearest neighbor classifier achieves a five-class classification accuracy of 97.40% on the Yaseen dataset and a binary classification accuracy of 89.11% on the PhysioNet/CinC Challenge 2016 dataset, providing valuable references for the prevention and diagnosis of cardiovascular diseases.
Last Update: 2026-07-23