Computers in biology and medicine · 2025
LSTM and reinforcement learning based delineation of electrocardiogram characteristic waves.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Merbouti Mohammed Abdenacer, Cherifi Dalila
Аннотация
BACKGROUND AND OBJECTIVE: Fiducial points of electrocardiogram (ECG) characteristic waves (CWs) are essential for clinical interpretation, yet manual delineation is labor-intensive and automated methods often depend on pre-segmentation or are sensitive to sampling variability. A particularly challenging case is P-wave delineation in third-degree atrioventricular (AV) block, where atrial and ventricular rhythms dissociate, complicating segmentation.
METHOD: This study proposes a two-stage framework combining a reinforcement learning (RL)-based peak analyzer with an LSTM classifier. The peak analyzer, tuned through RL exploration and exploitation ANNs, selects prominent peaks from normalized signals. These peaks are then classified by the LSTM into fiducial points (P, QRS, and T peaks and borders). The framework is designed to ensure robustness against predefined windows and signal resampling, address the challenge of P-wave delineation in third-degree AV block by incorporating rhythm- and morphology-related features, and improve computational efficiency through peak-based compression.
RESULTS: The peak analyzer reduced non-informative samples by 95.93 %, keeping delineation errors within 0-79 ms in time and 0-0.079 mV in amplitude. On intra-dataset tests, the best results were obtained on the CWD dataset, with sensitivities/PPVs of 96.85 %/95.97 % for peaks and 81.37 %/77.48 % for borders. On QTDB, performance reached 95.37 %/95.82 % for peaks and 80.66 %/79.97 % for borders, while on LUDB it achieved 85.04 %/87.04 % and 61.97 %/70.76 % respectively. In inter-dataset testing, training on LUDB and testing on QTDB yielded 80.84 %/84.06 % for peaks and 55.56 %/74.35 % for borders. Under Gaussian noise, performance degraded at 5 dB by -11.43 % sensitivity and -12.23 % PPV for peaks, and -11.99 %/-8.03 % for borders; at 10 dB, losses were -1.97 %/-8.24 % for peaks and -6.46 %/-7.47 % for borders.
CONCLUSION: The framework achieved precise, real-time delineation with a runtime of 40.11 ms per beat and reduced time error compared to related works. While P-wave delineation in third-degree AV block remains unresolved, the method provides robust and efficient performance, with potential applications in clinical ECG analysis and peak-based signal compression.
METHOD: This study proposes a two-stage framework combining a reinforcement learning (RL)-based peak analyzer with an LSTM classifier. The peak analyzer, tuned through RL exploration and exploitation ANNs, selects prominent peaks from normalized signals. These peaks are then classified by the LSTM into fiducial points (P, QRS, and T peaks and borders). The framework is designed to ensure robustness against predefined windows and signal resampling, address the challenge of P-wave delineation in third-degree AV block by incorporating rhythm- and morphology-related features, and improve computational efficiency through peak-based compression.
RESULTS: The peak analyzer reduced non-informative samples by 95.93 %, keeping delineation errors within 0-79 ms in time and 0-0.079 mV in amplitude. On intra-dataset tests, the best results were obtained on the CWD dataset, with sensitivities/PPVs of 96.85 %/95.97 % for peaks and 81.37 %/77.48 % for borders. On QTDB, performance reached 95.37 %/95.82 % for peaks and 80.66 %/79.97 % for borders, while on LUDB it achieved 85.04 %/87.04 % and 61.97 %/70.76 % respectively. In inter-dataset testing, training on LUDB and testing on QTDB yielded 80.84 %/84.06 % for peaks and 55.56 %/74.35 % for borders. Under Gaussian noise, performance degraded at 5 dB by -11.43 % sensitivity and -12.23 % PPV for peaks, and -11.99 %/-8.03 % for borders; at 10 dB, losses were -1.97 %/-8.24 % for peaks and -6.46 %/-7.47 % for borders.
CONCLUSION: The framework achieved precise, real-time delineation with a runtime of 40.11 ms per beat and reduced time error compared to related works. While P-wave delineation in third-degree AV block remains unresolved, the method provides robust and efficient performance, with potential applications in clinical ECG analysis and peak-based signal compression.