Colorectal Polyp Detection and Localization in Colonoscopy Images Based on YOLOv8

Our work focuses on determining the effectiveness of the YOLOv8 model in identifying colon polyps using the Kvasir dataset. Our research illustrates the YOLOv8 model’s exceptional performance throughout several epochs (50, 75, and 100), attaining elevated accuracy, recall, and mean Average accuracy (mAP) metrics. The model achieved a maximum mAP@0.5 of 99.3%, demonstrating its remarkable proficiency in polyp identification with an accuracy rate of 99.51%. The model achieved an F1-score of 98.8%, indicating an equitable balance between accuracy and recall. Comprehensive investigation demonstrated that augmenting training epochs improved detection accuracy, with optimal results noted between 50 to 75 epochs. The confusion matrix confirmed the extent to which the model could tell apart the different types of polyps thus placing YOLOv8 as a reliable tool in medical image analysis. This data implies that, depending on the final analysis, the YOLOv8 model can boost the quality and effectiveness of the first steps aimed at locating the polyp suggestion of the use of automated medical computer-aided systems.