PolySegNet: improving polyp segmentation through swin transformer and vision transformer fusion.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Colorectal cancer is a leading cause of cancer mortality globally.
• Colonoscopy is the primary diagnostic tool, but early polyp detection remains challenging.
• Automated polyp segmentation in colonoscopy images offers a promising solution.

Purpose of the Study:

• To develop and evaluate PolySegNet, an advanced deep learning model for accurate polyp segmentation in colonoscopy images.
• To leverage the power of Vision Transformer, Swin Transformer, and CNNs for enhanced polyp detection.
• To establish a new benchmark for transformer-based segmentation models in medical image analysis.

Main Methods:

• Proposed PolySegNet architecture integrating Vision Transformer and Swin Transformer with a CNN decoder.
• Extensive evaluation on three colonoscopy datasets, a combined dataset, and their augmented versions.
• Performance assessment using standard segmentation metrics like Dice score and Intersection over Union (IoU).

Main Results:

• PolySegNet achieved a mean Dice score of 0.92 and a mean IoU of 0.86.
• Demonstrated superior performance in accurately delineating polyp boundaries compared to existing methods.
• Exhibited competitive accuracy and efficacy in polyp segmentation across diverse datasets.

Conclusions:

• PolySegNet shows significant promise for accurate and efficient polyp segmentation in medical imaging.
• The proposed transformer-based architecture represents a potential foundation for future medical image analysis tools.
• This advancement could aid in earlier and more effective colorectal cancer diagnosis and prevention.