Real time Detection and Localization of Colorectal Polyps from Colonoscopy Images: A Deep Learning Approach

The growth of malignant colon polyps is a serious threat to both men’s and women’s lives, and image analysis performed during a colonoscopy is the procedure that is utilized the most frequently to test for benign and malignant polyps. Even though colorectal cancer is a major problem on a global scale, the mortality rate can be significantly lowered with the early diagnosis of polyps. This study made use of a model for the early detection of polyps. The model began by gathering data from the Kvasir-SEG dataset, and then proceeded to apply data augmentation techniques in order to boost the effectiveness of the dataset. The data were divided as follows: 80% percent for training and 20% for testing, including 20% of the training data utilized for validation to prevent overfitting. The study examined the dataset with YOLOv5 (You Only Look Once), specifically three variants of the model called YOLOv5s, YOLOv5m, and YOLOv5l. This allowed them to monitor each model’s performance while contrasting the results of the various models.. Results showed that YOLOv5l had the highest average testing IoU of 86.25% compared to the other employed models.