Leveraging YOLOv9 for Advanced Colorectal Polyp Detection

The increasing prevalence of colorectal cancer has necessitated improved diagnostic tools, which has spurred significant research efforts into Artificial Intelligence (AI)-assisted polyp detection and localization methods. Missed diagnoses due to human factors, such as fatigue or inexperience, are recognized to have severe consequences. This study investigates the efficacy of state-of-the-art object detection models for enhanced polyp identification, focusing on the performance of four variants of the YOLOv9 model (gelan-e, gelan-c, yolov9-c, and yolov9-e) for colorectal polyp detection and localization. These models were trained and tested using two distinct datasets: a combined dataset comprised of CVC-CLinicDB and Kvasir-SEG, and the LDPolypVideo dataset. The impact of different YOLOv9 architectures on detection accuracy and localization precision is analyzed. The YOLOv9 variants achieved mAP@50 scores up to 99.1% on CVC-ClinicDB (a 16% improvement over YOLOv8), outperforming YOLOv8 and other models, and 55.56% mAP@50 on LDPolypVideo, demonstrating enhanced accuracy and efficiency in colorectal polyp detection. This study highlights the potential of YOLOv9 to enhance the accuracy and efficiency of colorectal polyp detection.