Synthetic CT Pulmonary Angiography From Non-Contrast CT.

Computed tomography pulmonary angiography (CTPA) relies on iodinated contrast agents and poses risks including nephrotoxicity and allergic reactions. Our purpose is to develop a cascaded, tiny vessel-aware synthesis framework to generate high-fidelity synthetic contrast-enhanced CTPA (SyCTPA) from non-contrast CT (NCCT) scans. This study proposes a method to generate synthetic CTPA (SyCTPA) from non-contrast CT (NCCT) scans using a cascaded synthesizer based on Cycle-Consistent Generative Adversarial Networks (CycleGAN). A total of 410 retrospective paired CTPA and NCCT scans were obtained from three centers. The model was trained and validated internally on 249 paired images. An additional dataset comprising 161 paired images was used as a test set for model generalization evaluation and validation of downstream clinical tasks. The study was approved by the institutional review board, and all patient data were fully anonymized and de-identified prior to use. On the validation set, the proposed method achieved MAE of 156.28, PSNR of 20.71, and SSIM of 0.98, and on the independent test set, MAE of 165.12, PSNR of 20.27, SSIM of 0.98, and average CNR of 2.578. While the proposed method showed comparable pixel-level image similarity to pix2pix and standard CycleGAN, it demonstrated superior small-vessel enhancement in qualitative subtraction analysis and downstream vascular tasks. The approach was further applied to downstream tasks of pulmonary vessel segmentation and vascular quantification. On the test set, the average Dice, clDice, and clRecall of pulmonary segmentation using SyCTPA were 0.70, 0.71, and 0.73 for arteries and 0.70, 0.72, and 0.75 for veins, respectively, all improved compared with NCCT inputs. Intraclass correlation coefficient (ICC) for vessel volume between SyCTPA and CTPA was higher than that between NCCT and CTPA (average ICC, 0.81 vs 0.70), indicating effective vascular enhancement in SyCTPA, especially for small vessels. This study demonstrates the feasibility of synthesizing CTPA-like images from NCCT using a cascaded CycleGAN framework, offering a potential contrast-free approach to mitigate the risks associated with iodinated contrast agents in CTPA while preserving some diagnostic utility for pulmonary vascular diseases. Further validation, including radiologist reader studies and additional clinical endpoints, is required before clinical deployment.