Cardiovascular Anomaly Detection with Novel Approach
The cardiovascular system makes coordination of the circulation through all systems in the human body. Electronic signals collected from the human heart is an important research topic to explore cardiovascular anomalies in the body. The effective detection requires dynamic and robust classification mechanism even in the hardest environmental factors. The proposed study provides a Lightweight CNN-based solution using electrical activity signals of heart to categorize cardiovascular anomalies. This study synthesizes heterogeneously coherent ECG signals and adapts them into the input system of wearable devices. It consists of novel operations which are called Wave Form Geneticization, ECG Segment Shifting, Time Contraction and Expansion, Magnitudization, Projection, Permutation, Upscaling, Downscaling, Noising, and Denoising on the signal pattern to ensure a robust anomaly detection mechanism in such a sensor-independent and tempered manner. The proposed Wave-Segment Synthesizing method for heterogeneous harmonization of ECG signals differs from the literature by its robust classification capability of anomalies even in the weakest, loudest, most unstable, and corrupt data collection environments. It aims to provide a novel mechanism for wearable technologies with low computational capability. Thus, it will be more accessible to follow the disorders in the circulatory system and make an early diagnosis before death. It mitigates the vulnerabilities of literature models that can be affected by the deterioration of environmental factors such as sea level, pressure, optical effects, altitude and temperature. The applied dynamic algorithm is a candidate to be a pioneer for detecting abnormalities in the heart which closely concern all body systems in the human body.
Neural Networks Application for Heartbeat ECG for Real-Life Problem
Normal training process takes days. Below demo is just for quick demonstration of workflow of the neural network codes that constructed from scratch (without any framework or library like tensorflow):