A Transformer Based on Feedback Attention Mechanism for Diagnosis of Coronary Heart Disease Using Echocardiographic Images

Abstract

Coronary artery disease is a highly lethal cardiovascular condition, making early diagnosis crucial for patients. Echocardiograph is employed to identify coronary heart disease (CHD). However, due to issues such as fuzzy object boundaries, complex tissue structures, and motion artifacts in ultrasound images, it is challenging to detect CHD accurately. This paper proposes an improved Transformer model based on the Feedback Self-Attention Mechanism (FSAM) for classification of ultrasound images. The model enhances attention weights, making it easier to capture complex features. Experimental results show that the proposed method achieves high levels of accuracy, recall, precision, F1 score, and area under the receiver operating characteristic curve (72.3%, 79.5%, 82.0%, 81.0%, and 0.73%, respectively). The proposed model was compared with widely used models, including convolutional neural network and visual Transformer model, and the results show that our model outperforms others in the above evaluation metrics. In conclusion, the proposed model provides a promising approach for diagnosing CHD using echocardiogram.