Metody detekce arytmií z neinvazivního plodového elektrokardiogramu

Abstract

The dissertation deals with the design and validation of a processing pipeline for the automated assessment of fetal rhythm from non-invasively recorded abdominal ECG, including fetal signal extraction, fQRS complex detection, and subsequent analysis of rhythm abnormalities. The decision-making stage was based on the evaluation of the derived fetal heart rate, rhythm variability, and RR interval irregularity, enabling both binary screening of rhythm regularity and an extended analysis of suspicious recordings. The methodology was primarily validated using data from the NIFEA database, where binary classification achieved an overall accuracy of 84.6 %, a sensitivity of 91.7 % for the detection of irregular rhythm, and a specificity of 78.6 % for the recognition of regular rhythm. The extended analysis showed that, in suspicious recordings, the proposed approach is able to determine the predominant character of the rhythm abnormality, particularly ectopic, tachycardic, or bradycardic patterns. The proposed processing pipeline was further validated on real clinical data, which supported its practical applicability under conditions of lower and variable signal quality, while also highlighting the limitations associated with clinically recorded signals. The achieved results indicate that the proposed approach may serve as a tool for automated support in the screening of fetal rhythm abnormalities and as a basis for further development of non-invasive fetal monitoring methods, including potential applications in home monitoring and telemedicine.