Aortic Aneurysm Detection in CT Medical Images Based on YOLOv5

With the widespread application of computed tomography (CT) imaging technology, the detection of aortic dilation has become an important clinical diagnostic tool. This study proposes an improved model based on YOLOv5 for the automatic detection of aortic dilation in CT images. To address the insufficient feature extraction of low-contrast structures and the multi-scale fusion issues in the original YOLOv5 model when applied to medical imaging, ResNet50 is introduced as the backbone network. Its deep residual structure enhances the recognition of vascular walls and calcified plaques. Additionally, a dynamic weighted bidirectional feature pyramid network (BiFPN) is used to replace PANet, achieving adaptive feature fusion of multi-scale vascular structures. Experiments conducted on a dataset consisting of 9,856 CT images show that the improved model achieves an [email protected] of 88.79%, a 6.62% increase compared to the baseline model, while maintaining real-time performance. Notably, for small lesions (diameter < 10mm), the recall rate improved by 19.3%. Although the computational complexity of the model increased, it still meets the requirements for real-time clinical detection. The experimental results validate the effectiveness of the improved model in detecting aortic dilation, providing new optimization ideas and application potential for object detection in medical imaging.