A Real-Time Polyp Detection Framework for Colonoscopy Video

Colorectal cancer is one of the most common malignant tumors in the world. Endoscopy is the best screening method for colorectal cancer, which uses a micro camera to enter the colorectal and check whether there are polyps on the internal mucosa. In order to assist doctors to work more accurately and efficiently, a real-time polyp detection framework for colonoscopy video is proposed in this paper. The swin transformer block is integrated into the CNN-based YOLOv5m network to enhance the local and global information of the feature map. Then, in order to reduce the influence of factors such as light changes and reflection, we use the ensemble prediction of time series to improve the temporal continuity of the detection results. The experimental results show that compared with the baseline network, the precision rate of our method is improved by 5.3% and the recall rate is improved by 3.5%. And compared with recent research, our method achieves a good trade-off between detection speed and accuracy.