Brief Review — Synthesis of Normal Heart Sounds Using Generative Adversarial Networks and Empirical Wavelet Transform
GAN for Normal Heart Sound Synthesis
Synthesis of Normal Heart Sounds Using Generative Adversarial Networks and Empirical Wavelet Transform
GAN for Normal Heart Sound Synthesis, by Universidad del Norte
2020 MDPI J. Appl. Sci. (Sik-Ho Tsang @ Medium)Heart Sound Classification
2013 … 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 ====
- GAN is proposed to generate normal synthetic heart sounds.
Outline
- GAN for Normal Heart Sound Synthesis
- Results
1. GAN for Normal Heart Sound Synthesis
- (It is assumed that GAN is already known well.)
- Noise: a gaussian noise with a size of 2000 samples is used as input to the generator. The mean and standard deviation of the noise’s distribution are 0 and 1, respectively.
Generator model: It begins with a dense layer with ReLU activation function, followed by three convolutional layers with filters of size 128, 64, and 1, respectively; each of these layers have ReLU activation function, a kernel size of 3, and a stride of 1. Finally, there is a dense layer with a tanh activation function.
Discriminator model: It begins with a dense layer with ReLU activation function, followed by four convolutional layers with filters of size 256, 128, 64, and 32, respectively; each of these layers uses Leaky ReLU activation function, a kernel size of 3, and a stride of 1; additionally, between each convolution layer there is a Dropout of 0.25. Finally, there is a dense layer with a tanh activation function.
- Dataset of heart sounds: 100 normal heart sounds obtained from the Physionet database were used, with a sampling frequency of 2 KHz and 1 s of duration.
- Optimization: the Adam optimizer was used.
- Loss function: a binary, cross-entropy function was used.
2. Results
2.1. Visualization
It was observed that from 2000 epochs, the synthetic signal has a shape very similar to a natural signal, but with a relatively high noise level, as shown in Figure 8E. Therefore, it was decided to generate the synthetic signals up to 2000 epochs.
2.2. Mel-Cepstral Distortions (MCD)
- MCD is a measure of the difference between two MFCC sequences:
- A set of 400 natural normal heart sounds taken from the PhysioNet and PASCAL databases were used. Each signal was cut to a single cardiac cycle, with a normalized duration of 1 s, applying a resampling on the signal. Signals were also normalized in amplitude, and those signals with a similar heart rate were chosen.
- Those natural signals are compared to a total of 50 synthetic heart sounds.
The proposed method obtains small MCDs meaning that the synthesized signal is realistic, close to the real heart sound.
2.3. Classification Performance
- Models were trained with a total of 805 heart sounds (415 normal and 390 abnormal), obtained from the following databases: the PhysioNet/Computing in Cardiology Challenge 2016 [30], PASCAL challenge database [31], Database of the University of Michigan [48], Database of the University ofWashington [49], Thinklabs database (digital stethoscope) [50], and 3M database (digital stethoscope) [51].
- 50 synthetic signals were used as the test dataset.
From these results, it can be argued that the synthetic signals generated with the proposed method have similar characteristics to the natural signals, since the classification results on both type of signals are similar.
