Congenital Heart Disease Prediction based on Hybrid Approach of CNN-GRU-AM

Abstract

Chronic heart failure, also known as Congestive Heart Failure (CHD, is characterized by incapacitating symptoms that lead to higher rates of mortality and morbidity as well as higher medical costs and a lower quality of life. Detectable alterations on an Electrocardiogram (ECG) may be indicative of CHF using a simple and noninvasive diagnostic technique. The monitoring of cardiac patients with the use of heart signals has the potential to significantly increase life expectancy. For the past decade, patients and doctors have placed a premium on being able to classify and predict cardiac illnesses based on ECG data. Preprocessing, feature extraction, and model training were the three stages via which the research was conducted. Preprocessing often employs adaptive filters based on an LMS, however this can be time-consuming because of the filter’s long critical path. This issue is addressed by implementing a novel adaptive filter that makes use of a delayed error normalized LMS algorithm to achieve high speed and low latency. The preprocessed signal undergoes R-peak identification using wavelets for HRV feature extraction, and the resulting model is trained using these features CNN -GRU-AM. The experimental findings showed that compared to the CNN model (94%) and GRU (92%) model, the proposed model was significantly more accurate at 99.8%.

Topics & Concepts