Dual-source computed tomography image information under deep learning algorithm in evaluation of coronary artery lesion in children with Kawasaki disease

To explore the diagnostic value of deep learning algorithm in coronary arteries in children with Kawasaki disease, convolutional neural network (CNN) was applied in the segmentation of dual-source computed tomography (CT) (DSCT) image information of coronary artery lesions in children with Kawasaki disease. A CNN algorithm was improved, and the range image U-net (RIU-Net) was adopted to segment the CT image for convolution feature extraction. The improved model was fused with squeeze and excitation module and pyramid pooling to fabricate a new network (RISEU-Net), in which the ability of image features extraction was enhanced. Ninety children with coronary artery lesion of Kawasaki disease were selected and rolled into a control group (n = 45) and observation group (n = 45). The constructed model was used to segment the CT images of the two groups of patients, and the accuracy, sensitivity, and Dice of image segmentation were compared. The results showed that the utilization of key image features was improved after the SE module and pyramid pooling were integrated on the RIU-Net network. The lowest LOSS value of the RISEU-Net network was stable at about 0.02, and the effect was very fast during the training process. The sensitivity and accuracy of the constructed model to image detection were 95.6% and 96.7%, respectively. The negative predictive rate was 97.3%, and the positive predictive rate of 92.7%, which were all higher than those in the control group. The Dice values of RISEU-Net-N, RISEU-Net, and RIU-Net were 0.7765, 0.8562, and 0.8356, respectively. In summary, the RIU-Net network model proposed showed reliable feasibility and effectiveness for DSCT image segmentation, which effectively improved the clinical information evaluation of CT images of coronary artery lesions in children with Kawasaki disease, showing crucial reference for the development of intelligent medical equipment.