Polyp Detection from Colorectum Images by Using Attentive YOLOv5

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• High-quality colonoscopy is crucial for colorectal cancer prevention.
• Colonoscopy data primarily consists of images, necessitating advanced image analysis.
• Artificial intelligence (AI) in colonoscopy is a key research area for improving adenoma detection rates.

Purpose of the Study:

• To develop an AI model for accurate polyp detection during colonoscopy.
• To enhance the early detection of colorectal polyps and cancers.

Main Methods:

• A YOLOv5 model incorporating a self-attention mechanism was proposed for polyp target detection.
• The model utilizes image regression and attention mechanisms to focus on relevant features.
• Feature extraction is optimized to enhance informative channels and reduce interference.

Main Results:

• The AI model accurately identifies polyps, including small and low-contrast ones.
• Significant improvements in polyp detection speed were observed compared to existing algorithms.
• The method demonstrates high efficacy in real-time colonoscopy analysis.

Conclusions:

• The AI-assisted colonoscopy system significantly reduces missed diagnoses during endoscopic procedures.
• This technology holds great significance for improving clinical practice and patient outcomes.
• The study highlights the potential of AI to augment clinician capabilities in detecting colorectal polyps.