Application of dual branch and bidirectional feedback feature extraction networks for real time accurate positioning of stents

The installation of arterial stents refers to the use of stents (also known as vascular stents) to maintain the patency of arteries during the treatment of arterial stenosis or blockage. Arterial stents are typically made of metal or polymer materials and are structured as a mesh that provides support within the blood vessel, preventing it from collapsing again after interventional treatment. The installation of arterial stents is an effective interventional therapy that can significantly improve symptoms caused by arterial stenosis or blockage and enhance the quality of life for patients. Endovascular therapy has become increasingly important for treating both thoracic and abdominal aortic diseases. A critical aspect of this procedure is the precise positioning of stents and complete isolation of the pathology. To enhance stent placement accuracy, we propose a deep learning model called the Double Branch Medical Image Detector (DBMedDet), which offers real-time guidance for stent placement during implantation surgeries. The DBMedDet model features a parallel dual-branch edge feature extraction network, a bidirectional feedback feature fusion neck sub-network, as well as a position detection head and a classification head specifically designed for thoracic and abdominal aortic stents. The model has achieved a detection Mean Average Precision (mAP) of 0.841 ([email protected]) and a real-time detection speed of 127 Frames Per Second (FPS). For [email protected], when employing 5-fold cross-validation, DBMedDet demonstrates superior performance compared to several YOLO models, achieving improvements of 4.88% over YOLOv8l, 4.61% over YOLOv8m, 3.20% over YOLOv8s, 6.23% over YOLOv8n, 6.09% over YOLOv10s, 3.92% over YOLOv9s, 3.20% over YOLOv8s, 3.00% over YOLOv7tiny, and 5.01% over YOLOv5s. This study presents a precise and easily implementable method for the automatic detection of stent placement limits in the thoracic and abdominal aorta. The model can be applied in various areas such as coronary intervention therapy, peripheral vascular intervention therapy, cerebrovascular intervention therapy, postoperative monitoring and follow-up, and medical training and education. By utilizing real-time imaging guidance and deep learning models (such as DBMedDet), stent placement procedures in these application areas can be performed with greater precision and safety, thereby enhancing patient treatment outcomes and quality of life.