RFPNet: Reorganizing feature pyramid networks for medical image segmentation.

Area of Science:

• Medical Imaging
• Computer Vision
• Artificial Intelligence

Background:

• Accurate medical image segmentation is vital for clinical treatment planning.
• Challenges include data acquisition difficulties and lesion tissue heterogeneity.
• Existing methods struggle with automatic and precise segmentation.

Purpose of the Study:

• To propose a novel network, Reorganization Feature Pyramid Network (RFPNet), for improved medical image segmentation.
• To address the challenges of automatic and accurate segmentation in diverse clinical scenarios.
• To construct semantic features at various scales and levels using cascaded Thinned Encoder-Decoder Modules (TEDMs).

Main Methods:

• RFPNet comprises base feature construction, feature pyramid reorganization, and multi-branch feature decoding modules.
• TEDMs are alternately cascaded to build multi-scale semantic features.
• The network reorganizes multi-level features and recalibrates feature channels.

Main Results:

• RFPNet achieved high Dice scores (e.g., 90.47% on ISIC2018) and Jaccard scores (e.g., 83.95% on ISIC2018) across multiple datasets (ISIC2018, LUNA2016, RIM-ONE-r1, CHAOS).
• The proposed method demonstrated superior performance compared to classical and state-of-the-art segmentation techniques.
• Visual results confirmed RFPNet's excellent capability in segmenting target areas in clinical images.

Conclusions:

• RFPNet offers a robust solution for accurate medical image segmentation.
• The network's architecture effectively handles multi-scale feature representation and integration.
• RFPNet shows significant potential for clinical applications requiring precise image segmentation.