Phonocardiogram classification based on 1D CNN with pitch-shifting and signal uniformity techniques
View/Open
Date
2024Author
Ahmad, ZafarKhan, Muhammad Salman
Chowdhury, Muhammad E.H.
Zughaier, Susu
Ibrahim, Wanis Hamad
Metadata
Show full item recordAbstract
This study combines 1D CNN with advanced signal processing to enhance heart sound classification, presenting three key contributions. Initially, we used a pitch-shifting technique to expand the dataset by altering high-frequency components precisely, ensuring the preservation of vital information. Next, a signal normalization technique is deployed, equalizing signal lengths for uniform analysis across all samples. Utilizing 1D CNN and Mel-frequency cepstral coefficients (MFCCs) for feature extraction, our approach achieves notable classification accuracy, with results showing up to 99.57% accuracy, 99.80% specificity, 99.22% sensitivity, and a 99.22% F1 score. These developments not only advance the precision of heart sound classifications but also expand the potential for wider clinical applications, establishing a new benchmark in tele auscultation.
Collections
- Electrical Engineering [2811 items ]
- Medicine Research [1633 items ]