Increasing Reliability of Detecting Bleed Sites in Angiograms• Investigated a two-stage deep learning based object detection approach (bleeding classifier and bleeding localizer) for identifying and detecting bleed sites in angiograms to enable more effective embolization• Performed data augmentation operations using imgaug due to the limited size of the dataset• Trained Faster R-CNN model through ChainerCV to generate bounding boxes around bleeding spots over patients’ X-ray images and the model performance achieved an accuracy (IoU) of 79%