RAC-Net: Interpretable Medical Small Target Segmentation Network with X-ray Radiation Attenuation Characterization

Area of Science:

• Medical Imaging
• Computer Vision
• Radiology

Background:

• Medical image segmentation is crucial for diagnosis and treatment.
• Current methods struggle with small targets due to poor feature extraction and background interference.

Purpose of the Study:

• To develop a novel network (RAC-Net) for enhanced small medical target segmentation.
• To improve early lesion detection and localization accuracy.

Main Methods:

• Proposed Radiation Attenuation Characteristics Network (RAC-Net) inspired by X-ray attenuation.
• Integrated Radiation Attenuation Salient Feature Extraction (RASFE) module for feature enhancement and noise suppression.
• Incorporated Partial Volume Artifact Edge Enhancement (PVA) module using CT Hounsfield Unit priors to address feature distortion and blurring.

Main Results:

• RAC-Net achieved state-of-the-art performance in medical small target segmentation.
• Demonstrated significant improvements with 18% DICE and 19% IOU gains over existing methods.
• Validated enhanced robustness and accuracy through extensive experiments.

Conclusions:

• RAC-Net effectively boosts small target segmentation accuracy in medical imaging.
• The network provides robust support for early disease diagnosis and intervention.
• The proposed modules successfully address limitations in sparse feature extraction and artifact handling.