Y-Polyp: research on devices for detecting colorectal polyps with limited samples

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Colorectal polyps can become cancerous; early detection via colonoscopy is key to reducing cancer incidence and mortality.
• Computer-aided detection (CAD) systems using neural networks show promise for polyp identification.
• Existing CAD systems struggle with limited training data, hindering performance.

Purpose of the Study:

• To introduce the Y-Polyp model for improved colorectal polyp detection.
• To address limitations of insufficient training data in current CAD systems.

Main Methods:

• The Y-Polyp model utilizes a parallel strategy with multidimensional attention mechanisms.
• This approach allows convolutional kernels to learn flexible attention across four spatial dimensions, capturing features from limited data.
• The model filters conflicting spatial information and inconsistencies across different feature scales.

Main Results:

• The Y-Polyp model demonstrated significant improvements in colorectal polyp detection accuracy.
• Precision increased by 3%, recall by 2.7%, and average precision by 2% and 2.7% respectively.
• The model effectively handles limited data samples.

Conclusions:

• The Y-Polyp model exhibits strong generalization and robustness across various colon structures and polyp appearances.
• This colorectal polyp detection approach is highly applicable and reliable.
• The Y-Polyp model offers a promising solution for computer-aided polyp diagnosis.