HSSAM-Net: hyper-scale shifted aggregation network for precise colorectal polyp segmentation in endoscopic images.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Colorectal cancer is a leading cause of mortality, necessitating early detection via colonoscopy.
• Accurate polyp segmentation in endoscopic images is crucial but challenging due to image variability and artifacts.
• Current methods struggle with precise polyp identification, impacting early diagnosis.

Purpose of the Study:

• To develop a lightweight deep learning framework, HSSAM-Net, for accurate and efficient polyp segmentation in colonoscopy images.
• To improve multi-scale contextual information capture, feature propagation, and texture representation for enhanced segmentation.
• To provide a computationally efficient solution for real-time clinical applications in computer-aided colonoscopy.

Main Methods:

• Proposed HSSAM-Net framework integrating Hyper-Scale Shifted Aggregation Module (HSSAM) and Progressive Reuse Attention.
• Incorporated Max-Diagonal Pooling/Unpooling (MaxDP/MaxDUP) for improved texture representation and feature alignment.
• Evaluated on five benchmark datasets: Kvasir, CVC-ClinicDB, ETIS, CVC-300, and EndoCV2020.

Main Results:

• HSSAM-Net achieved state-of-the-art accuracy with Dice scores of 0.949-0.952 and mIoU of 0.924-0.930.
• The model demonstrated real-time efficiency at 24.1 FPS with only 0.9 million parameters.
• Consistent outperformance of state-of-the-art methods across multiple benchmark datasets.

Conclusions:

• HSSAM-Net offers a favorable balance between accuracy and computational efficiency for polyp segmentation.
• The model's performance and speed make it suitable for real-time clinical applications in computer-aided colonoscopy.
• HSSAM-Net advances the development of practical and reliable systems for early colorectal cancer detection.