LightAttn-YOLO-V8: an efficient colorectal polyp detection with lightweight attention mechanism based on YOLO

Colorectal cancer has a high rate of mortality globally, often originating from polyps that, if detected early, can significantly improve patient outcomes. This study proposes an improved polyp detection method by incorporating a lightweight attention mechanism into the YOLOv8 framework, resulting in a model referred to as LightAttn-YOLO-V8. Our methodology is crafted to address the complexities of polyp detection in endoscopic images, where discrepancies in size, form, and appearance hinder precise identification. By integrating both channel and spatial attention mechanisms, our model focuses on essential features while ensuring computational efficiency, making it suitable for real-time clinical use. We trained and assessed our model on a comprehensive dataset that includes Kvasir-SEG, CVC-ClinicDB, ETIS, and CVC-ColonDB, yielding outstanding outcomes. On the ETIS dataset, our model achieved a recall of 93.9%, a precision of 97.6%, and an F1-score of 95.7%, whereas on CVC-ClinicDB, it reached a flawless recall of 100%, a precision of 99.3%, and an F1-score of 99.6%. These findings surpass many leading-edge methods, illustrating the capacity of our approach to improve early polyp detection and assist clinicians in alleviating the impact of colorectal cancer. Our research emphasizes the effectiveness of attention mechanisms in enhancing detection precision without significantly compromising speed, thereby paving the way for more reliable computer-aided diagnosis in endoscopy.