Brief Review — Abnormal Heart Rhythm Detection Based on Spectrogram of Heart Sound using Convolutional Neural Network
Spectrogram+CNN
Abnormal Heart Rhythm Detection Based on Spectrogram of Heart Sound using Convolutional Neural Network
Spectrogram+CNN, by STIKOM Bali, Universitas Pendidikan Ganesha
2018 CITSM (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]
==== My Other Paper Readings Are Also Over Here ====
- Heart sound is converted into spectrogram as input to CNN for normal/abnormal classification.
Outline
- Spectrogram+CNN
- Results
1. Spectrogram+CNN
1.1. Spectrogram
- Total abnormal and normal audio data are 480 data. The amount of normal heartbeat sound are 351 data and the amount of abnormal heartbeat sound are 129 data.
Before being processed using the Convolutional Neural Network, audio data is converted to two-dimensional image first. Heartbeat sound data are represented by spectrogram in two-dimensional form.
1.2. CNN
A CNN is proposed, which has 2 convolutional layers, and 2 dense layers as above.
2. Results
Training accuracy tends to increase as the number of epoch increases. Meanwhile, testing accuracy rate didn’t increase significantly. This condition called overfitting.
The classification model in the 50th epoch was the best model, it detects unlabeled data more than another classification model with different epoch number.
