Survey for Detecting Heart Diseases

Sik-Ho Tsang

·

Follow

3 min read

·

May 11, 2024

--

Listen

Share

A Comprehensive Survey of Analysis of Heart Sounds using Machine Learning Techniques to Detect Heart Diseases
Survey for Detecting Heart Diseases, by Sri Sathya Sai University for Human Excellence Gulbarga, RV University Bangalore
2023 JCTCP (Sik-Ho Tsang @ Medium)

Phonocardiogram (PCG)/Heart Sound Classification
2013 … 2023 … [CTENN] [Bispectrum + ViT]
==== My Other Paper Readings Are Also Over Here ====

• An estimated 32 per cent of all global deaths were due to cardiovascular diseases (CVD) in 2019. Of these, three-fourths of the deaths occur in low and middle-income nations.
• A survey has been studied for machine learning (ML) and deep learning (DL) techniques to detect the most common heart diseases.
• (There are many overview or survey papers for heart sound classification. This is one of them. And authors mention some of the research gap at the end of paper.)

Outline

1. Survey
2. Research Gaps

1. Survey

1.1. Pre-processing

• Denoising and segmentation might be performed.

1.2. Feature Extraction

• Fourier Transform, Shannon enery envelop, Discrete wavelet transforms (DWT), continuous wavelet transforms (CWT), Cepstrum, Bispectrum, Wigner bispectrum and Mel Frequency Cepstrum Coefficient (MFCC), are also used for feature extraction in the literature.
• The features extracted are then fed to the ML algorithms for classification.

1.3. ML & DL for Classification

• The traditional ML techniques that have been used for heart sounds analysis include SVMs, k-NN and RF, among others.
• Some of the researchers using DL techniques have skipped the segmentation phase. Among DL, DNN, CNN, and RNN are used.
• CNN can help in identifying subtle changes in heart sounds.
• RNN can detect the subtle changes in audio.
• Most of them work on PhysioNet dataset.
• (In the paper, each reference is described by 1–3 sentences. Please read the paper for more details if interested.)

2. Research Gaps

• Limited availability of labelled data: Too few dataset.
• Heterogeneity of heart sounds: Heart sounds vary slightly between individuals and within the same individual over time.
• Lack of standardisation in data acquisition and processing: Standardised protocols are being developed for segmentation and classification, to promote consistency.
• Interpretability: The AI models are difficult to be understood and interpreted due to their black box nature.
• TinyML and Edge Computing: TinyML involves deploying ML algorithms on small, portable devices. Edge computing, on the other hand, is the processing of the data locally instead of the cloud.