Automatic Detection and Classification of Cardiovascular Disorders Using Phonocardiogram and Convolutional Vision...

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Cardiology

Background:

• Cardiovascular disorders (CVDs) are a leading global cause of mortality.
• Existing deep learning (DL) systems for CVD diagnosis lack optimal accuracy and require substantial computational resources and data.
• There is a need for efficient and accurate diagnostic tools for CVDs.

Purpose of the Study:

• To develop an accurate and computationally efficient deep learning model for diagnosing cardiovascular disorders (CVDs) using phonocardiogram (PCG) signals.
• To introduce a novel attention-based convolutional vision transformer (CVT-Trans) for PCG signal analysis.
• To address the limitations of current DL-based CVD diagnostic systems.

Main Methods:

• A novel attention-based convolutional vision transformer (CVT-Trans) architecture was developed.
• Continuous Wavelet Transform-based Spectrogram (CWTS) was employed for feature extraction from PCG signals.
• The CVT-Trans model was trained to categorize PCG signals into five distinct classes.

Main Results:

• The CVT-Trans system achieved an overall average accuracy of 100%, sensitivity (SE) of 99.00%, specificity (SP) of 99.5%, and F1-score of 98%.
• Performance was validated using 10-fold cross-validation, demonstrating robustness.
• The proposed method significantly outperformed existing state-of-the-art techniques.

Conclusions:

• The developed CVT-Trans technique offers a highly accurate and robust solution for CVD diagnosis.
• This cost-effective approach, utilizing PCG signals, can aid cardiologists in diagnosing heart valve problems.
• The model's high performance suggests its potential for widespread clinical application.