Skip to main content

Table 6 Studies for HS feature extraction and classification using machine learning

From: Prediction of exercise sudden death in rabbit exhaustive swimming using deep neural network

Year Author Dataset Feature extraction methods Classifier Results
Sens (%) Spec (%) Acc (%)
2015 Zheng et al. [23] 88 normal heart sounds, 64 abnormal heart sounds MF-DFA, MESE, EMD HMM 82.95 79.68 81.58
BP-ANN 85.23 82.81 84.21
LS-SVM 96.59 93.75 95.39
2016 Thomae et al. [29] PhysioNet 1D CNN Bidirectional GRU 96 83
2016 Potes et al. [43] PhysioNet LR-HMMS, MFCC AdaBoost 70 88
Frequency bands decomposition CNN 79 86
2019 Li et al. [25] 2532 recordings from healthy subjects, 664 recordings from patients DAE 1D CNN 97.85
MFCC 1D CNN 91.02
2020 Li et al. [24] PhysioNet Eight domains CNN 87 86.6 86.8
2020 Gao et al. [27] 1286 normal recordings form PhysioNet, 108 abnormal heart sounds from patients SVM 87.62
FCN 94.65
LSTM 96.29
GRU 98.82
2020 Deng et al. [28] PhysioNet MFCC CRNN 98.66 98.01 98.34
PRCNN 97.33 97.33 97.34
  1. MF-DFA multifractal detrended fluctuation analysis, MESE maximum entropy spectra estimation, EMD empirical mode decomposition, CNN convolutional neural network, 1D CNN one-dimensional convolutional neural network, DAE denoising autoencoder, MFCC Mel-frequency cepstrum coefficient, HMM hidden Markov model, LR-HSMM logistic regression-based hidden semi-Markov model, BP-ANN back-propagation artificial neural network, LS-SVM least square support vector machine, GRU gated recurrent unit, FCN Fully Convolutional Network, LSTM long-short term memory network, CRNN convolutional recurrent neural networks, PRCNN paralleling recurrent convolutional neural network