Návrh algoritmu pro detekci R-vrcholů založeného na umělé inteligenci

Abstract

This bachelor thesis focuses on the issue of automatic R-peak detection in electrocardiogram (ECG) signals using artificial intelligence methods. The main objective is the design, training, and testing of four neural network architectures within the MATLAB environment. This group includes a base and a deep convolutional neural network (CNN), a bidirectional long short-term memory (BiLSTM) network, and their hybrid combination. The MIT-BIH NSRDB database was utilized for training, with the data being formatted into a three-channel input to increase detection robustness. The proposed models were tested on an independent annotated dataset and subsequently compared with conventional detectors. The results showed that all proposed models achieve high performance, with F1-scores exceeding 0.999, thereby outperforming selected conventional methods. The base 1D CNN proved to be particularly effective and an ideal choice for this type of task. The thesis also includes the development of a software application that enables the visualization of ECG signals and the practical application of the developed models. This work confirms that deep learning represents a highly effective tool for robust R-peak detection.