Brief Review — Towards classifying non-segmented heart sound records using instantaneous frequency based features


Sik-Ho Tsang
4 min readJan 28, 2024
PCG Signals

Towards classifying non-segmented heart sound records using instantaneous frequency based features
, by Yarmouk University
2019 TandF IJMT, Over 50 Citations (Sik-Ho Tsang @ Medium)

Heart Sound Classification
2023 [2LSTM+3FC, 3CONV+2FC] [NRC-Net] [Log-MelSpectrum+Modified VGGNet] [CNN+BiGRU] [CWT+MFCC+DWT+CNN+MLP] [LSTM U-Net (LU-Net)]
==== My Other Paper Readings Are Also Over Here ====

  • The heart sound features are extracted for the classification of non-segmented signals. The method has two major phases:
  • The first phase is to estimate the instantaneous frequency of the recorded signal.
  • The second phase is to extract a set of 11 features from the estimated instantaneous frequency.


  1. IFE+RF, IFE+kNN
  2. Results


1.1. Instantaneous frequency estimation (IFE)

Instantaneous Frequency Estimation (IFE)
  • The definition of instantaneous frequency of a nonstationary signal as defined by boualem that it is a time-varying parameter that related to the mean of the frequencies in the signal as it develops [26, 27].
  • The power spectrum (time-frequency distribution) Ps(s,t) of X(t) is computed.
  • Then, the instantaneous frequency of X(t) is computed as below:
  • The above figure shows an example of IFE over spectorgram of the signal.
Binary Class
  • The above figure shows IFE alongside with spectrogram for normal and abnormal signals.
  • As we can notice from both IFE there is a difference between the two IFE which reflected on the extracted statistical feature from them and on the performance of the classifiers.

1.2. Feature Extraction

  • After estimating the instantaneous frequency for each PCG signal, these signals are used to extract 11 statistical features.
  • These features are maximum by mean, maximum by a median, standard deviation by mean, mean, standard deviation, maximum, minimum, median, a1, a3, and a4 features:
  • Principal component analysis (PCA) is also used for dimensionality reduction, feature ordering, and feature selection.

1.3. Classification

  • Random forest (RF) and kNN are used as classifier.

2. Results

2.1. Multiclass IFE

IFE Examples for 5 Classes

2.2. RF & kNN Performance

RF & kNN Performance

Both classifiers (RF and KNN) achieve high accuracy in the classification of normal and abnormal heart sounds on binary dataset.

RF outperforms kNN on multiclass dataset.

2.3. Performance Using PCA

Performance Using PCA
  • The features selection start from 3 features with an increment of two features index each time (3, 5, 7, 9, and 11 features).
  • RF classifier with binary dataset PCA can reduce the features by 18% to 9 features instead of 11 while for the kNN classifier with the same dataset it can reduce the features by 72% to three features only.
  • For multiclass dataset, the PCA can reduce the features set used with RF classifier by 18% to 9 features with performance a little lower than 11 features and by 18% for kNN classifier with the highest best performance.

2.4. SOTA Comparisons

SOTA Comparisons
  • There are both segmented and non-segmented SOTA approaches in the above table.

The proposed system shows that the new provided IFE based statistical features have high classification rates compared to other methods, the higher classification rates over other methods in the literature using both the RF and KNN classifiers.



Sik-Ho Tsang

PhD, Researcher. I share what I learn. :) Linktree: for Twitter, LinkedIn, etc.