Capsule Network for 1-D Biomedical signals: A Review

Abstract

The heartbeat, muscle contractions, and other phys- iological functions are examples of biomedical signal sources. Electrocardiograms (ECG), electroencephalograms (EEG), and electromyograms (EMG) are examples of the signals that can be non-invasively recorded and used for diagnosis and as health in- dicators. Hence, timely and accurate diagnosis of the biomedical signals plays a prominent role. Professional healthcare workers assess the signal in search of a clear pattern that would indicate a normal or abnormal heartbeat is a tedious job. Manual inter- pretation of the signals may lead to misdiagnosis. The automated computer-aided diagnosis (CAD) method is one way to support decision-making for the eradication of these deficiencies. The CAD tool should operate as a real-time system for early diagnosis, requiring little time investment, data dependence, and device- specific measurement variances. Deep learning-based methods are becoming more and more common in CAD techniques. Convolutional neural network (CNN), one of the well-known deep learning network, fail of recognise position, texture, and genetic anomalies in the image. A capsule network is one of the newest and most promising deep learning algorithms that tackles CNN’s shortcomings. In this study, we present a thorough analysis of the cutting-edge methodology, tools, and topologies used in current capsule network implementations. The key contribution with this review study is its explanation and summary of major existing Capsule Network implementations and architectures.