| Abstract [eng] |
This dissertation presents the results of a multidisciplinary collaboration between researchers in cardiology and a bioengineering team, integrating novel wearable ECG technologies, clinical validation in patients with arrhythmia, the use of artificial intelligence for ECG analysis, and risk stratification for atrial fibrillation development. The research contributes to the advancement of digital health technologies and is based on four peer-reviewed scientific papers, multiple international patents and the MDR certification of the invented device. The author was a principal investigator responsible for the clinical evaluation of the invented device in Paper I and II as well as a co-author of the presented patents, Paper III and Paper IV. The Paper I reports the clinical validation of a wrist-worn device that combines continuous photoplethysmography (PPG) with intermittent six-lead wire-free ECG. The combined approach was clinically validated and shown to provide high diagnostic accuracy in patients with atrial fibrillation despite the presence of frequent premature beats in the control groups. The Paper II highlights the diagnostic value of six-lead standard-limb-like ECG compared to single lead-I-like ECG. The six-lead ECG demonstrated superior accuracy over both single lead-I-like ECG and PPG-only detectors in the presence of control groups with sinus rhythm and ectopic activity. The number of premature beats per minute proved to be the key factor associated with false positives versus true negatives across all diagnostic tools. Widespread use of smartwatch-recorded single-lead ECGs may substantially increase the risk of false-positive diagnoses (type I error) in populations with frequent premature contractions. In Paper III, the challenges of analyzing large-scale ECG data in ambulatory monitoring and telemedicine were addressed. The author annotated and coordinated the Lithuanian annotation panel during the data labeling process. In this extensive ambulatory ECG dataset, the deep learning model (DeepRhythmAI), designed for direct-to-physician reporting, outperformed manual annotations by board-certified technicians in detecting critical arrhythmias. In Paper IV, the identification of ECG-based atrial fibrillation risk markers within the first 24 hours of Holter monitoring enabled early risk stratification of patients likely to benefit from prolonged rhythm surveillance. This finding provides an additional approach to enhance patient selection. From the perspective of practical application, the thesis includes a case report of the first life saved by the invented wearable device. It enabled a timely diagnosis of a life-threatening paroxysmal atrial tachycardia that had previously remained undetected using conventional methods, ultimately leading to successful catheter ablation with long-term effectiveness. In addition, the author provided recommended practical applications of the invented device in various clinical scenarios. |
