ECG analysis and machine learning applications in healthcare diagnostics

Abstract

Atrial Fibrillation (AF) and Barrett's Esophagus (BE) are significant health conditions that require advanced diagnostic methods and risk stratification to optimize patient care. AF, the most common cardiac arrhythmia, increases the risk of stroke and heart failure. BE, a precancerous condition, can progress to esophageal adenocarcinoma, making careful monitoring essential to prevent cancer development. Current diagnostic methods for both conditions often need to be improved, such as low accuracy, insufficient risk assessment, and high operator dependence. This dissertation addresses these challenges by developing advanced diagnostic methodologies to enhance the detection and management of AF and BE.The first part focuses on ECG analysis techniques to detect and classify AF. AF is characterized by chaotic and complex electrical activity in the heart, which traditional ECG methods fail to capture fully. This research introduces modified time-delayed embedding techniques to distinguish paroxysmal AF from normal sinus rhythm. Additionally, the Complexity AF score integrates AF burden, electrical burden, and Poincaré analysis, offering a dynamic tool to assess AF severity and complexity. These approaches aim to improve early detection, accurate diagnosis, and personalized treatment by predicting outcomes. The second part introduces a deep learning (DL) framework to enhance BE surveillance and management. Existing methods, such as random biopsies and white light endoscopy, encounter challenges like sampling errors and difficulties in detecting non-visible (microscopic) recurrences, particularly at the squamocolumnar junction (SCJ). Advanced imaging techniques like optical coherence tomography (OCT) show promise in providing images at a microscopic level. This study uses DL-based analysis of OCT images to localize the SCJ and detect glandular structures indicative of dysplasia. Combining these methods with 3D esophagus reconstructions of the human esophagus identifies potential areas for BE recurrence, improving the precision and efficiency of BE monitoring, reducing unnecessary invasive diagnostics, and enabling targeted interventions. By integrating innovations in ECG signal processing and DL-based imaging techniques, this dissertation advances the detection, risk assessment, and management of AF and BE, offering potential improvements in clinical outcomes and personalized patient care.