Rethinking the detail-preserved completion of complex tubular structures based on point cloud: A dataset and a...

Area of Science:

• Medical Imaging
• Computer-Assisted Diagnosis
• Point Cloud Processing

Background:

• Complex tubular structures are vital for medical imaging and diagnosis.
• Current segmentation methods fail with discontinuities in cases like coronary artery stenosis, impacting accuracy.
• Reconnecting these structures is crucial for complete anatomical visualization.

Purpose of the Study:

• To address the challenge of discontinuous tubular structures in medical imaging.
• To introduce the first point cloud-based approach for tubular structure completion.
• To establish a new benchmark dataset (PC-CAC) for evaluating tubular structure completion methods.

Main Methods:

• Exploration of tubular structure completion using point clouds.
• Establishment of the Point Cloud-based Coronary Artery Completion (PC-CAC) dataset from clinical data.
• Proposal of TSRNet (Tubular Structure Reconnection Network) with a detail-preservated feature extractor, multiple dense refinement, and global-to-local loss.

Main Results:

• TSRNet demonstrates accurate reconnection while preserving structural integrity.
• Experiments on PC-CAC, PC-ImageCAS, and PC-PTR datasets show superior performance over existing methods.
• The proposed method sets a new benchmark for point cloud-based tubular structure reconstruction.

Conclusions:

• The developed PC-CAC dataset and TSRNet effectively address tubular structure discontinuities.
• The approach significantly improves downstream diagnostic accuracy in medical imaging.
• This work advances point cloud-based reconstruction for complex anatomical structures.