Image Binary Classification of Coronary Artery Stenosis Based on Resnet18 and Transfer Learning

Classification detection of coronary artery stenosis plays a vital role in assisting physicians to diagnose cardiovascular diseases. To classify the degree of coronary stenosis accurately and efficiently, CT angiography (CTA) images are converted into binary images by a joint segmentation method, which is based on maximum between-class variance (OTSU) and region growing. Further, the binarized datasets are normalized, which realizes the preprocessing of the medical datasets. Then, to enhance the classification performance, a classification model of coronary artery stenosis, which is based on ResNet18 and transfer learning, is constructed. Lastly, a dataset of CTA images from real patients is applied for experimental verification. The experiment shows that the accuracy of the proposed method is up to 96.97%, 8.08% higher than ResNet18 model, and 5.06% higher than the ResNet18 model with spatial attention mechanism and channel attention mechanism. It can be concluded that the efficiency and reliability of coronary stenosis classification can be considerably improved.