Tubular Structure Segmentation via Multi-Scale Reverse Attention Sparse Convolution.

Area of Science:

• Medical Imaging
• Computer Vision
• Biomedical Engineering

Background:

• Accurate segmentation of cerebrovascular and airway structures is crucial for morphological research and pathological detection.
• Complex morphological and topological characteristics of these tubular structures pose significant segmentation challenges.

Purpose of the Study:

• To develop a robust framework for accurate segmentation of cerebrovascular and airway tubular structures.
• To address the challenges posed by complex anatomical features and image artifacts in medical imaging.

Main Methods:

• Proposed a UARAI framework integrating a U-Net multi-scale reverse attention network and a sparse convolution network.
• Utilized multi-scale structures for global and detailed feature extraction, and a reverse attention module for fine-edged feature enhancement.
• Incorporated sparse convolution to improve feature expression without increasing model complexity and a cropping strategy to mitigate boundary effects.

Main Results:

• UARAI achieved Dice scores of 90.31% and 93.34% for cerebrovascular and airway segmentation, respectively.
• IoU scores reached 82.33% for cerebrovascular and 87.51% for airway segmentation.
• Demonstrated superior accuracy and robustness compared to existing segmentation techniques.

Conclusions:

• The UARAI framework offers a significant advancement in the accurate segmentation of tubular structures in medical images.
• The proposed method shows strong potential for improving medical image analysis, clinical diagnosis, and supporting healthcare professionals.