Multi-scale feature pyramid fusion network for medical image segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate medical image segmentation is vital for clinical diagnosis and research.
• Challenges include blurred borders and low contrast in Computed Tomography (CT) imaging.
• Existing methods often struggle with complex anatomical structures.

Purpose of the Study:

• To introduce a novel Multi-Scale Feature Pyramid Fusion Network (MS-Net) for improved medical image segmentation.
• To address limitations in segmenting organs with challenging visual characteristics in CT scans.
• To enhance the accuracy and reliability of diagnostic image analysis.

Main Methods:

• Developed MS-Net, a deep learning model integrating Multi-Scale Attention Module (MSAM) and Stacked Feature Pyramid Module (SFPM).
• MSAM extracts multi-level contextual details via dynamic receptive field adjustment.
• SFPM focuses network attention on target organs by adaptively weighting features.

Main Results:

• MS-Net significantly improved Dice scores on CHAOS (91.74% to 94.54%), Lung (97.59% to 98.59%), and ISIC 2018 (82.55% to 86.06%) datasets compared to U-Net.
• Outperformed six other state-of-the-art methods in metrics like Miou, Dice, ACC, and AUC.
• Demonstrated superior performance in segmenting challenging medical images.

Conclusions:

• Both MSAM and SFPM techniques contribute to improved segmentation efficacy.
• MS-Net achieves superior results in medical image segmentation tasks, including CHAOS, Lung, and ISIC 2018.
• The proposed network offers a promising advancement for clinical diagnostic imaging.