Arrhythmia Detection In ECG Signals Using Deep Learning Models

Abstract

As the technology is emerging every day, there is a crucial need of advanced medical diagnosis especially for cardiac diseases. In the real meaning of advanced healthcare, this study addresses the challenges, accurate and timely detection of cardiac disease arrhythmia to save life of multiple individuals. Arrhythmia that is irregular beats of heart that can cause severe life threatening complications like heart failure and attacks if not treated swiftly. The key objective of this research project is to design and implement a resilient framework using deep learning models on a MIT-BIH Arrhythmia Dataset. This system will handle a wide range of ECG signals at different variations like amplitude and duration to detect arrhythmia and its multiple types with increased diagnostic accuracy. Through this early detection would happen by an advanced solution to improve the patient care, clinical practice and avoid any kind of human error. In the methodology, deep learning models are deployed, including Convolution Neural Network (CNN), Long Short-Term Memory (LSTM), Dense Net and Sequential Model, and the accuracy of the model was optimized. To this end, a user-friendly graphical interface has also been created to ensure the ability of these models, thereby widening spread of timely detection. The results of this work show that the Sequential Neural Network model outperformed all other models by achieving an accuracy of 98% which overall increases the accuracy by 2.3% as compared to the conventional research. By improving the accessibility, accuracy, monitoring and elimination of human error the system provides better treatment outcomes, optimizes healthcare professional services and minimize administrative burden on healthcare professionals.