Using real-time image enhancement technique to improve prostate capsula detection

Plasmakinetic resection of the prostate (PKRP) and plasmakinetic enucleation of the prostate (PKEP) are two effective methods for treating benign prostate hyperplasia (BPH). During surgery, the doctors will trim away the extra prostate tissue using the boundary of the prostate capsula. When some doctors resect more, prostate capsula perforation will occur. At this time, accurate and real-time prostate capsula detection becomes extremely urgent. In addition, poor quality and low contrast images are the main problems in the medical field. After enhancement, the objective images are visibly more distinct when encountering image noise and clutter. Right now, popular object detection methods like YOLOv3 and Single Shot MultiBox Detector (SSD) can detect prostate capsule in real time, but the results are not ideal. We improve these methods using deep bilateral learning, which is a one real-time image enhancement technique, and the method increases the accuracy by 2 ~ 2.69 percent. We implement the experiments using YOLOv3, the SSD, YOLOv3 with deep bilateral learning, and the SSD with deep bilateral learning. The best detection results are provided by the SSD combined deep bilateral learning with an accuracy of 98.39 percent. The proposed method has the potential for specific practical uses in medical image diagnosis.