Two Projections Suffice for Cerebral Vascular Reconstruction

Area of Science:

• Medical Imaging
• Computer Vision
• Neuroscience

Background:

• 3D reconstruction of cerebral vasculature from 2D projections is crucial for diagnosis and treatment planning.
• Traditional methods face challenges due to inherent ambiguities in 2D projections, leading to unsatisfactory outcomes.

Purpose of the Study:

• To introduce a novel approach for accurate 3D cerebral vasculature reconstruction from 2D biplanar projections.
• To overcome the limitations of traditional backprojection methods by resolving ambiguities.

Main Methods:

• Utilized a U-Net architecture trained to resolve ambiguities in initial backprojections.
• Employed a Maximum A Posteriori (MAP) strategy with a continuity prior for refinement.
• Evaluated using a dataset of segmentations from ~700 MR angiography scans, generating realistic biplanar digitaly reconstructed radiographs (DRRs).

Main Results:

• Achieved 80% Dice similarity against ground truth on held-out data.
• Demonstrated superior reconstruction quality compared to existing methods.
• The U-Net and MAP strategy effectively resolved ambiguities and enhanced continuity.

Conclusions:

• The novel approach significantly improves the quality of 3D cerebral vasculature reconstructions.
• This method holds promise for enhanced diagnosis and treatment planning in neurovascular conditions.
• The developed code and dataset are publicly available for further research.