Real-Time Registration for Fast KV-Switching 2D Dual-Energy Angiography
E Nikolau*, M Wagner, P Laeseke, M Speidel, University of Wisconsin - Madison, Madison, WI
Presentations
TH-F-201-5 (Thursday, 7/14/2022) 2:00 PM - 3:00 PM [Eastern Time (GMT-4)]
Room 201
Purpose: To evaluate the performance of real-time registration techniques for fast kV-switching 2D dual-energy subtraction angiography in a continuously-ventilated large animal model.
Methods: Dual-energy pulmonary angiography was performed in three continuously-ventilated swine using an interventional C-arm system equipped with 30 Hz fast kV-switching capability. Iodine/bone-selective dual-energy images were generated using a weighted log subtraction of low/high kV image pairs. A real-time block-matching approach was implemented to estimate a displacement vector field and correct for inter-frame respiratory motion in the 1/30 s period between adjacent low/high energy projections. Block-matching registration was compared to diffeomorphic demons registration, which served as the reference method. Both methods were tested on a commercial GPU (3584 core, 11 GB). Normalized root-mean-square-error (nRMSE) relative to the reference was computed in regions of interest (n=270) in iodine/bone-selective dual-energy images. Additionally, the relative size and intensity of respiratory-induced motion artifacts over a single respiratory cycle were quantified in iodine/bone-selective images with and without application of either registration approach. Computation time per frame was measured to assess suitability for real-time application.
Results: nRMSE in dual-energy images without registration ranged from 19.8%-28.8% across all three swine. For the block-matching approach, nRMSE ranged from 10.7%-16.5%. The average reduction in diaphragm artifact profile intensity across all three swine was 61.7% using diffeomorphic demons registration, and 46.8% using block-matching. Similarly, artifact profile area was reduced by 61.8% using demons registration, and 40.8% using block-matching. Computation time was 21 ms/frame for block-matching and 51 ms/frame for diffeomorphic demons registration.
Conclusion: Both block-matching and diffeomorphic demons methods offer a reduction in artifacts caused by minor motion between high and low kV images. A tradeoff between diaphragm artifact reduction and computational speed was observed. On the tested hardware, block-matching is feasible for 30 fps real-time processing whereas diffeomorphic demons is more suitable for <20 fps processing.
Funding Support, Disclosures, and Conflict of Interest: Financial support was provided by Siemens Healthineers.
Keywords
Dual-energy Imaging, Angiography, Registration
Taxonomy
IM- X-Ray: Dual-energy and spectral
Contact Email