In Journal of vascular and interventional radiology : JVIR
PURPOSE : To investigate the utility and generalizability of deep learning subtraction angiography (DLSA) for generating synthetic digital subtraction angiograms (DSA) without misalignment artifacts.
MATERIALS AND METHODS : DSA images and native digital angiograms of the cerebral, hepatic, and splenic vasculature, both with and without motion artifacts, were retrospectively collected. Images were divided into a motion-free training set (n=66 patients, 9,161 images) and a motion artifact-containing test set (n=22 patients, 3,322 images). The deep neural network pix2pix was trained to produce synthetic DSA images without misalignment artifacts directly from native digital angiograms using motion-free DSA images of the cerebral and hepatic vasculature as ground truth. After training, the algorithm was tested on digital angiograms of hepatic and splenic vasculature with substantial motion. Four board-certified radiologists evaluated performance via visual assessment with a five-grade Likert scale. Subgroup analyses were performed to analyze the impact of transfer learning and the ability of this approach to generalize to novel vasculature.
RESULTS : Compared to the traditional DSA method, the proposed approach was found to generate synthetic DSAs with significantly fewer background artifacts (mean rating of 1.9 [95% CI 1.1-2.6] vs. 3.5 [95% CI 3.5-4.4]; p=0.01), without a significant difference in foreground vascular detail (mean rating of 3.1 [95% CI 2.6-3.5] vs. 3.3 [95% CI 2.8-3.8]; p=0.19) in both the hepatic and splenic vasculature. Transfer learning significantly improved the quality of generated images (p<0.001).
CONCLUSIONS : DLSA successfully generates synthetic angiograms without misalignment artifacts, is improved through transfer learning, and generalizes reliably to novel vasculature that was not included in the training data.
Crabb Brendan T, Hamrick Forrest, Richards Tyler, Eiswirth Preston, Noo Frederic, Hsiao Albert, Fine Gabriel C
2022-Dec-15
