Boundary-aware and discrepancy-guided dynamic pseudo-labeling with consistency learning for semi-supervised 3D...

Area of Science:

• Medical Imaging
• Neuroscience
• Computer Vision

Background:

• Cerebrovascular diseases pose significant global health risks, necessitating accurate diagnostic tools.
• Precise segmentation of cerebrovascular structures in Time-of-Flight Magnetic Resonance Angiography (TOF-MRA) is vital for diagnosis and treatment planning.
• Current segmentation methods face challenges due to anatomical variability and limited annotated data.

Purpose of the Study:

• To develop a semi-supervised framework for accurate cerebrovascular segmentation in 3D TOF-MRA.
• To address limitations in vessel morphology variability and annotation scarcity.
• To improve the efficiency and reliability of cerebrovascular segmentation for clinical applications.

Main Methods:

• Proposed BDD-CL (Boundary-Aware and Discrepancy-Guided Dynamic pseudo-labeling Consistency Learning) framework.
• Incorporated a Boundary Enhancement (BE) module for improved vessel boundary delineation using shape constraints.
• Integrated a Shape-Aware Discrepancy (SAD) module to enhance robustness in complex morphologies by refining prediction inconsistencies.
• Utilized a Dynamic Pseudo-label Selection (DPS) mechanism for adaptive pseudo-label generation, optimizing label efficiency and mitigating error propagation.

Main Results:

• BDD-CL demonstrated superior performance compared to seven state-of-the-art semi-supervised methods.
• Quantitative and qualitative evaluations on COSTA and IXI datasets confirmed the framework's effectiveness.
• The method achieved significant improvements in cerebrovascular segmentation accuracy.

Conclusions:

• The BDD-CL framework shows strong potential for label-efficient and reliable cerebrovascular segmentation in clinical practice.
• The proposed approach offers a promising solution for overcoming challenges in TOF-MRA segmentation.
• The study contributes to advancing automated analysis of cerebrovascular structures.