Real-time Volumetric Imaging for Minimally-Invasive Thoracic Interventions: A Deformable Phantom Study

VoxelMap reconstructs real-time volumetric respiratory motion from 2D C-arm projections for two deformable lung phantoms, achieving SSIM >0.92 and inference under 15 ms.

• Why it matters: Continuous volumetric guidance during minimally invasive thoracic procedures is limited by slow 3D imaging, creating a need for fast, reliable motion estimation from fluoroscopy.
• What they did: The method was trained on 4D CBCT and evaluated on withheld respiratory phases for PixelPrint4D and Casper phantoms using SSIM/PSNR, measured inference time, and deep ensembles for uncertainty.
• The result: SSIM exceeded 0.92 in all experiments (Casper 0.969–0.973; PixelPrint4D 0.927–0.943), inference was <15 ms, and ensemble uncertainty correlated positively with prediction error in both phantoms.

AI summary of the abstract · may contain errors · not medical advice · read the paper.