Improved polyp detection from colonoscopy images using finetuned YOLO-v5

Object detection plays an important role to accelerate the medical diagnosis and automatic polyp detection from colonoscopy images is one of the prominent examples. Visual examination is an error prone and time-consuming process to determine the shape, size, and location of polyps in colonoscopy images. Deep learning models are extensively accepted in the field of object detection. YOLO-v5 is an object detection model which is powered by deep learning technique. In this study we suggest a finetuned YOLO-v5 model for polyp detection from colonoscopy images. Effective data augmentation is also done to improvise the performance of the model. To quantify the efficiency of our solution against other works, two aspects have been fundamentally considered - qualitative performance and practical reliability, which have been achieved successfully. A detailed comparison study has been shared to justify the performance of our proposed method against other deep learning techniques - R-CNN, Faster RCNN and YOLO-v4. Our proposed solution deliberated high value of performance metrics which are comparatively better than other solutions.