Deep Learning Model for Spectrogram-Based Cardiac Abnormality Detection on Mobile Fuzzy-Net

Cardiovascular Diseases are a leading cause of global mortality, necessitating rapid and accurate detection of cardiac abnormalities. Electrocardiogram analysis reveals heart electrical activity, and spectrogram representation enhances diagnostic accuracy. However, traditional Short-Time Fourier Transform-based spectrograms lack resolution and suffer from spectral leakage, failing to capture fine-scale frequency variations crucial for diagnosing certain heart diseases. This study proposes a deep learning method based on Mel-Frequency Cepstral Coefficients (MFCC) spectrograms, which combine Mel-Filter bank and Discrete Cosine Transform to enhance frequency resolution and reduce spectral leakage. Mobile-Net facilitates efficient feature extraction without compromising diagnostic performance. The research examines deep learning techniques to predict four major cardiac abnormalities: abnormal heartbeat, myocardial infarction, history of myocardial infarction, and normal heart rhythm. Transfer learning with pre-trained deep neural networks and a new convolutional neural network (CNN abnormality prediction. To further enhance performance and address ECG data uncertainty, a novel Mobile Fuzzy-Net (MFN) integrating Mobile-Net with Adaptive Neuro-Fuzzy Inference Systems (ANFIS) is introduced. Experimental results indicate that the MFN model surpasses existing works, by achieving 98.22% recall, 98.31% precision, and a 98.21% F1 score, with the integration of MFN for feature extraction in Mobile-Net and ANFIS achieving a peak score of 99.79%. These results underscore the potential of combining advanced signal processing, deep learning, and fuzzy logic techniques to create robust and precise cardiac abnormality detection systems, enhancing clinical diagnosis and patient outcomes.) with Mobile-Net architecture are explored for cardiac