Click here to

Session: CAD, Radiomics, and Quantitative Applications of X-Ray Imaging and CT [Return to Session]

Motion-Compensated Quantitative Digital Subtraction Angiography (qDSA) in Noisy Image Sequences

J Whitehead*, S Periyasamy, C Hoffman, P Laeseke, M Speidel, M Wagner, University of Wisconsin - Madison, Madison, WI


TH-B-207-2 (Thursday, 7/14/2022) 8:30 AM - 9:30 AM [Eastern Time (GMT-4)]

Room 207

Purpose: Motion-compensated quantitative digital subtraction angiography (MC-qDSA) has been proposed for blood velocity quantification from interventional DSA image sequences in which respiratory or cardiac-related motion causes motion of the arteries. Due to the high frame rate (30 frames/s) required for qDSA, velocity measurements from low dose per frame images are being investigated. This work studied the accuracy of vessel motion-compensation at increased noise levels associated with dose reduction.

Methods: MC-qDSA uses a deep-learning, frame-by-frame vessel segmentation followed by vessel registration to track vessel centerlines, which are subsequently used as measurement points to compute blood velocity. To characterize the sensitivity of the MC-qDSA centerline calculation to image noise encountered at decreased radiation dose levels, continuously ventilated hepatic angiograms from two swine were acquired at 3.6μGy/frame, and zero-mean correlated noise was added to simulate angiograms acquired at 1.8, 0.8, and 0.2μGy/frame. Vessel centerline measurements were made in three vessels for each swine, in 150 frames per subject. Accuracy of MC-qDSA vessel centerlines was evaluated using the average shortest distance (dₑ) of all points from the manually-labeled reference vessel centerlines to the motion-compensated centerlines and vice versa. The impact of centerline variations on time-attenuation curve (TAC) measurements used in the MC-qDSA algorithm were evaluated by extracting TACs from the 3.6μGy/frame image sequence and computing the normalized root-mean-squared-difference (nRMSD) between the results obtained with MC and manual centerlines.

Results: Mean±SD dₑ at each dose level, averaged over all vessels, was 3.3±2.3mm without motion-compensation and 0.3±0.3mm with motion-compensation. The corresponding TAC nRMSD was 32±9% without motion-compensation and 11±4% with motion-compensation.

Conclusion: In swine model images with noise levels representative of 0.2-3.6 μGy/frame detector dose rate, MC-qDSA centerline placement was consistent to within 0.3±0.3mm. Results support the feasibility of motion-compensated qDSA at reduced radiation dose rates relative to standard DSA at 3.6 μGy/frame.

Funding Support, Disclosures, and Conflict of Interest: Funding was received from Siemens Healthineers through a sponsored research agreement


Angiography, Flow Velocity, Noise


IM- X-Ray: Quantitative imaging

Contact Email