A METHOD TO IMPROVE THE ACCURACY OF DEEP LEARNING MODELS IN CARDIOVASCULAR DISEASE CLASSIFICATION USING ELECTROCARDIOGRAM SIGNALS

Tóm tắt

This paper proposes an improved method to enhance the accuracy of classifying cardiovascular diseases based on electrocardiogram signals by applying a deep learning model composed of multiple integrated components. Specifically, the model architecture is built upon a one-dimensional convolutional neural network to extract local features from raw electrocardiogram signals, effectively capturing significant patterns in the input data. Subsequently, a long short-term memory network is employed to exploit the temporal dependencies within the signal, enabling the model to understand contextual relationships and dynamic changes in features over time. To further improve the model's ability to focus on the most relevant information for classification, a multihead attention mechanism is integrated after the long short-term memory layer. This attention mechanism allows the model to learn the relative importance of different segments within the signal sequence more effectively. Experimental results demonstrate that the combination of one-dimensional convolutional neural network, long short-term memory, and multihead attention yields high performance, achieving an accuracy of over 97% in classifying four types of heart diseases. The proposed method shows promising potential for the application of artificial intelligence in the automated diagnosis of cardiovascular conditions.