Research on the deep learning study of left ventricle CT segmentation

Cardiovascular illnesses are becoming a major global health concern requires sophisticated, accurate and efficient cardiac computed tomography heart image analysis, to assist diagnosis and therapy planning, particularly accurate left ventricle segmentation. In order to overcome this, this paper created an enhanced U-Net model that greatly enhances left ventricular detail identification by combining a novel multi-scale feature fusion with an attention mechanism. By comparing the performance on the same dataset, the enhanced U-Net shows relative advantages in important metrics such as precision, recall, F1 score, IoU and Dice coefficient. The enhanced U-Net model is more effective in acquiring knowledge and more reliable in reaching a steady state, which is significant for practical clinical applications. Furthermore, the model is optimised by using depthwise separable convolution, which guarantees lightness and increases training speed. Extensive experiments on many cardiac CT datasets validated the model's exceptional segmentation accuracy, computational performance, and flexibility, outperforming state-of-the-art deep learning techniques and offering new opportunities for accurate diagnosis and treatment of cardiovascular diseases.