Advanced QT interval analysis in long-term electrocardiography using shape-based clustering and template matching: A novel approach for Holter monitoring.

Introduction: The QT interval, a critical parameter in electrocardiography (ECG), is challenging to analyze accurately because of variations in T-wave morphology, especially in long-term ECG recordings, such as Holter monitoring. Current methods, which are often manual or semi-automated, lack consistency and efficiency, underscoring the need for a more reliable and automated approach. Methods: We developed a novel QT interval analysis algorithm that integrates K-shape clustering with Dynamic Time Warping under Limited Warping Path Length for template matching. This method automatically categorizes ECG waveforms into clusters based on shape similarity, and then selects representative template waveforms for each cluster. This approach aimed to standardize and automate the process of identifying and analyzing QT intervals across varying waveform patterns. Results: Validation of the algorithm using the QT Database, which encompasses a broad spectrum of ECG waveform patterns, confirmed its ability to accurately identify and analyze the QT interval across diverse T-wave morphologies, in close agreement with expert human analysis. The reliability of the algorithm was quantified using the Intraclass Correlation Coefficient. For the QT interval, the Intraclass Correlation Coefficient (2,1) was 0.9 [95 % CI: 0.89, 0.91], and for the QTcB interval, it was 0.82 [95 % CI: 0.80, 0.84], both reflecting high reliability. The reliability of the algorithm was also evaluated by comparing it with DSC5500 from Nihon Kohden analysis program for long-time ECGs, a commercial software program that uses the tangential method. For the QT interval calculated by the developed algorithm, the Intraclass Correlation Coefficient (2,1) was 0.95 [95 % CI: 0.92, 0.97], and for the QT interval calculated by the DSC 5500, it was 0.70 [95 % CI: 0.54, 0.81]. For all waves, the mean and standard deviation of the difference between the algorithm and the expert analysis was 0.5 (±16), and that of the difference between the DSC5500 and the expert analysis was -16.4 (±42.8). For biphasic T-wave, the mean and standard deviation of the difference between the algorithm and the expert analysis was -0.3 (±15.9), and that of the difference between the DSC5500 and the expert analysis was -79.3 (±60.6). The results showed that the developed algorithm performed better than the DSC5500 for all shapes, especially for biphasic T-wave. The algorithm also demonstrated enhanced time efficiency and streamlined QT interval analysis in clinical settings. Discussion: The new algorithm enhanced automated QT interval analysis, especially for long-term ECG recordings. Its effectiveness in processing different T-wave morphologies with accuracy and efficiency makes it a promising tool for clinical use and improving QT interval analysis in patient care and diagnostics.