S1 and S2 Heart Sound Recognition Using Deep Neural Networks

Area of Science:

• Cardiology
• Biomedical Engineering
• Signal Processing

Background:

• Accurate identification of heart sounds (S1 and S2) is crucial for diagnosing cardiac conditions.
• Traditional methods often rely on timing intervals and durations, which can be challenging to obtain reliably.
• Exploring acoustic characteristics independently offers a potential alternative for robust heart sound recognition.

Purpose of the Study:

• To investigate the feasibility of recognizing S1 and S2 heart sounds solely based on acoustic features.
• To evaluate the performance of a deep neural network (DNN) approach in heart sound recognition without temporal assumptions.
• To determine if reliable recognition is achievable when duration and interval information is inaccessible.

Main Methods:

• A deep neural network (DNN) model was developed for S1 and S2 heart sound recognition.
• Heart sound signals were converted into Mel-frequency cepstral coefficients (MFCCs).
• K-means clustering refined MFCC features, which were then classified by the DNN.

Main Results:

• The proposed DNN-based method achieved high precision, recall, and F-measure scores.
• The system demonstrated an accuracy rate exceeding 91% for S1 and S2 heart sound recognition.
• The DNN classifier outperformed other established pattern classification methods.

Conclusions:

• Reliable S1 and S2 heart sound recognition is achievable using only acoustic characteristics.
• The DNN method offers a robust approach, independent of ECG or temporal assumptions.
• This acoustic-feature-based recognition holds promise for simplified and effective cardiac auscultation.