Brief Review — Complexity measures of high oscillations in phonocardiogram as biomarkers to distinguish between normal heart sound and pathological murmur
DWT for Feature Extraction + SVM for Classification
Complexity measures of high oscillations in phonocardiogram as biomarkers to distinguish between normal heart sound and pathological murmur
DWT+SVM, by Concordia University, European University Institute, Wilfrid Laurier University, and IPAG Business School
2022 J. Chaos, Solitons and Fractals (Sik-Ho Tsang @ Medium)Heart Sound Classification
2013 … 2022 [CirCor Dataset] [CNN-LSTM] [DsaNet] [Modified Xception] [Improved MFCC+Modified ResNet] [Learnable Features + VGGNet/EfficientNet] 2023 [2LSTM+3FC, 3CONV+2FC] [NRC-Net]
==== My Other Paper Readings Are Also Over Here ====
- DWT is used to transfrom time domain heart sound signal to wavelet domain. Then various features are extracted from the wavelet signal.
- SVM is used to classify whether the heart sound is normal or having murmur.
Outline
- DWT+SVM
- Results
1. DWT+SVM
1.1. DWT Feature Extraction
- Discrete wavelet transform (DWT) is used to transform sound signal x.
- The discrete form of low frequency oscillations v and high frequency oscillations w at scale j are respectively denoted by Sj and Dj given by:
In this work, the daubechies-4 wavelet function is applied for heart sound record analysis at third level of decomposition.
- The subsequent nonlinear patterns (HE, LZC, SE) which are described next are all estimated from DWT-based detail coefficients Dj.
- Hurst Exponent (HE), Lempel-Ziv Complexity (LZC), and Shannon Entropy (SE) are used as features fed into SVM.
- (Please read the paper directly for more details of HE, LZC and SE.)
1.2. SVM Classification
- Non-linear SVM is used.
- The parameters of the nonlinear SVM classifier are tuned by using Bayesian optimization.
2. Results
2.1. Dataset
- PhysioNet Dataset is used.
2.2. Results
The proposed approach outperforms [17] for each set in PhysioNet Dataset.
