Classification of Atrial Fibrillation with Pre-Trained Convolutional Neural Network Models

Atrial Fibrillation (AF) is the most common chronic arrhythmia. Effective detection of the AF would avoid serious consequences like stroke. Conventional AF detection methods need heuristic or hand-crafted features extraction. Recently, deep learning (DL) techniques with massive data have been used on image, voice and other field widely with impressive results. The ECG rhythms such as AF, normal, Noisy and other rhythms have been segmented into 3 segments per rhythm, converted into 2D images using short time Fourier transform (STFT) and fed into pre-trained models. The pre-trained CNN models are used for transfer learning or are fine tuned for the detection and classification of the AF rhythm. The features extracted from the last layer of the pre-trained models are used as input to classical classification algorithms such as Ensemble classifier and support vector machine (SVM) for AF detection. The proposed approach would be have great potential on real-time monitoring of atrial fibrillation signal in electrocardiogram. Overall, our approaches achieved accuracy, sensitivity and specificity as of 97.89%, 97.12% and 96.99% similar to the latest state of the art techniques but with more flexibility.