Mohammad Sarfraz, Mudassir Hasan Khan, Duraid Yahya Mohammed, Mays Dheya Hussain & Khamis A. Zidan
الملخص الانجليزي
With the evolving lifestyle, many cardiac ailments are becoming more frequent, and it has become necessary to provide detailed surveillance of the heart's functioning to ensure healthy living. ECG signals provide details regarding the various forms of arrhythmias. However, owing to the complexities and non-linearity of ECG signals, it is impossible to examine these signals manually. Conventional approaches for specialist inspection of ECGs on paper or television are inadequate for ambulatory, long-term monitoring, and sports ECGs. Automated applications that use signal processing and pattern recognition would be extremely beneficial. Identification of arrhythmias from ECGs is an essential branch of biomedical signal processing and pattern recognition. Motion-induced artifacts are well-known to be a major source of misrecognition and misdiagnosis.
On the other hand, the feature extraction method has a significant effect on the reliability and performance of ECG pattern recognition. This paper proposes new approaches and algorithms for pre-processing multi-channel ECG signals and neural networks for arrhythmia classification using independent component analysis (ICA) with two distinct goals: (1) to eliminate motion-induced or associated artifacts, and (2) to better select the features and allow more effective pattern recognition. When processing noisy ECG data with the MIT dataset, cross-validation reveals a major improvement. For noisy signals, classification sensitivity of 97.9% and positive predictivity of 98.1% was achieved in this work. A tenfold neural validation rule was used to achieve 99.3% accuracy in arrhythmia classification. The lower the signal-to-noise level, the more significant is the improvement. This proposed algorithm would be a valuable method for physicians in justifying their diagnosis. With its quick reaction time, the proposed algorithm can be easily integrated into an automated healthcare management system.
أ د خميس عواد زيدان | Khamis A. Zidan | 10857