Branch and End Points Detection in Cerebral Vessels Images Using Deep Learning Object Detection Techniques

In this study, we introduce a cutting-edge methodology for detecting branching and endpoints in two-dimensional brain vessel images, employing deep learning-based object detection techniques. While conventional image processing methods are viable alternatives, our adoption of deep learning showcases notable advancements in accuracy and efficiency. Following meticulous cleaning and labeling of the raw dataset sourced from laboratory environments, we meticulously convert it into the COCO format, ensuring compatibility with deep learning algorithms for both training and testing phases. Utilizing four deep learning object detection methods: fast R-CNN, faster R-CNN, RetinaNet and RPN within the Detectron2 framework, our study achieves remarkable results. Evaluation using the intersection over union (IoU) method underscores the robust performance of our deep learning approach, boasting a success rate surpassing 90%. This breakthrough not only enhances neuroimaging analysis but also holds immense potential for revolutionizing diagnostic and research practices in neurovascular studies.