Research on automatic analysis and diagnosis system of cardiovascular images based on deep learning

Cardiovascular disease (CVD) is a disease with high morbidity and mortality worldwide. Traditional diagnostic methods have some problems such as strong subjectivity and limited accuracy. With the development of medical imaging technology, cardiovascular imaging data are increasingly rich, but how to analyze these data efficiently and accurately has become an urgent problem. In this study, an automatic analysis and diagnosis system of cardiovascular images based on deep learning (DL) is developed and verified, the system is designed to enhance the precision and speed of diagnosis, utilizing convolutional neural networks (CNN) as the central feature extraction component, and realizes multi-task learning of lesion classification, vessel segmentation and stenosis regression in cardiovascular images through multi-scale feature fusion network combined with pre-trained VGG16 model. The experiment used CT and MRI image data of 1000 patients, the data was split into three sets: training, verification, and test, with a distribution ratio of 8:1:1. The results indicated that the model achieved an accuracy of 93.2%, along with corresponding recall and F1 scores, 88.5% and 90.7% for mild stenosis, respectively. Moderate stenosis was 90.5%, 91.2% and 90.8%; Severe stenosis was 91.7%, 92.3% and 92.0%. Dice coefficient is stable at 0.85 after 100 rounds of training, which shows that the model has high accuracy in the task of blood vessel segmentation. The average absolute error of stenosis regression task is 5%, which verifies the effectiveness of the model in dealing with complex medical image regression tasks.