TransResU-Net: A Transformer based ResU-Net for Real-Time Colon Polyp Segmentation

Area of Science:

• Medical imaging analysis
• Artificial intelligence in healthcare
• Gastroenterology

Background:

• Colorectal cancer (CRC) is a leading cause of cancer mortality globally.
• Timely colon cancer screening is vital for early detection, but current colonoscopy methods have a significant polyp miss rate.
• Early polyp detection can reduce CRC mortality and healthcare costs.

Purpose of the Study:

• To develop a deep learning-based computer-aided diagnosis (CADx) system for automatic polyp segmentation.
• To improve the polyp detection rate during colonoscopy and aid in early CRC diagnosis.
• To establish a benchmark for real-time polyp detection systems.

Main Methods:

• Proposed a novel deep learning architecture named Transformer ResU-Net (TransResU-Net).
• The architecture integrates residual blocks (ResNet-50 backbone), transformer self-attention, and dilated convolutions.
• Evaluated the model on two public polyp segmentation datasets.

Main Results:

• TransResU-Net achieved a highly promising dice score, indicating effective segmentation accuracy.
• The model demonstrated real-time processing speed, suitable for clinical application.
• Experimental results suggest high efficacy in polyp detection metrics.

Conclusions:

• TransResU-Net shows significant potential as a benchmark for real-time polyp detection systems.
• This AI system can assist gastroenterologists in identifying missed polyps, improving early CRC diagnosis.
• The developed system contributes to cost-effective, long-term colorectal cancer prevention.