A semantic feature enhanced YOLOv5-based network for polyp detection from colonoscopy images.

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Gastroenterology

Background:

• Colorectal cancer (CRC) presents high global morbidity and mortality.
• Current computer-assisted colonoscopy for polyp detection faces challenges like missed or false detections.
• Improving polyp detection accuracy is crucial for effective colonoscopy.

Purpose of the Study:

• To propose an improved YOLOv5-based method for accurate colorectal cancer polyp detection.
• To enhance feature expression and focus on polyp characteristics in colonoscopy images.
• To address the limitations of existing polyp detection algorithms.

Main Methods:

• Incorporation of a novel P-C3 structure into the YOLOv5 backbone and neck networks.
• Introduction of a contextual feature augmentation module with a coordinate attention mechanism.
• Utilizing an improved YOLOv5 architecture for enhanced multi-scale feature analysis.

Main Results:

• The proposed method demonstrates significant advantages in polyp detection accuracy compared to traditional algorithms.
• The improved model shows a notable increase in recall rate, effectively reducing missed polyp detections.
• Experimental results validate the model's superior performance in identifying polyps.

Conclusions:

• The developed YOLOv5-based method significantly improves polyp detection rates in colonoscopy.
• This advancement contributes to reducing missed diagnoses and enhancing the effectiveness of colorectal cancer screening.
• The study holds importance for clinical practice and the development of endoscopic diagnostics.