MCDNet: Morphological-conditional dual-view fusion for 3D tubular structure segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer Vision

Background:

• Accurate segmentation of 3D tubular structures is vital for medical diagnosis and treatment planning.
• Current deep learning models struggle with generalizability across different anatomical regions due to reliance on specific morphological priors.
• Joint modeling of global and local morphological characteristics in tubular structure segmentation is underexplored.

Purpose of the Study:

• To develop a novel deep learning network, MCDNet, for generalizable 3D tubular structure segmentation.
• To integrate both contextual and morphological information effectively for improved segmentation performance.
• To address the limitations of existing methods in handling diverse tubular geometries and anatomical variations.

Main Methods:

• Proposed MCDNet (Morphological-Conditional Dual-view Network) incorporating a target-adaptive Morphological-Conditional Convolution (MCConv).
• Implemented a three-stage architecture: morphological feature extraction, contextual feature learning with cross-fusion, and residual self-attention fusion.
• MCConv enhances structural sensitivity across diverse tubular shapes, while cross-fusion combines convolutional and attention-based representations.

Main Results:

• MCDNet achieved superior performance on four diverse benchmark datasets for tubular segmentation.
• Demonstrated an average Dice coefficient improvement of 6.93% compared to a strong baseline.
• Showcased a 10.61% reduction in Hausdorff distance, indicating improved boundary accuracy.

Conclusions:

• MCDNet offers a robust and generalizable solution for 3D tubular structure segmentation in medical imaging.
• The integration of morphological and contextual features via MCConv and dual-view architecture is key to its improved performance.
• The proposed method advances the state-of-the-art in medical image segmentation for complex anatomical structures.