Department of Radiological Technology, Tohoku University Hospital, 1-1 Seiryo-machi, Aoba-ku, Sendai, Miyagi, 980-8574, Japan , hirokazu.takano.a1@tohoku.ac.jp
Abstract: (7 Views)
Background: Single-energy computed tomography (SECT) requires different energies for disease diagnosis and detection. The patient must be scanned again after the CT scan to obtain other tube-voltage images, causing increased radiation exposure. One method for avoiding rescanning is virtual monochromatic images (VMI) by dual-energy computed tomography (DECT). However, VMI has not proven superior to SECT. Moreover, DECT is not available at all facilities. This study used generative adversarial networks to generate 120 kVp and Sn140 kVp CT images from 80 kVp CT images. Material and Methods: This study involved 35 patients with pulmonary hypertension who underwent CT scans with DECT. The peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) were calculated to evaluate the difference between the real and pseudo images. Furthermore, the mean CT values of the pulmonary arteries (PA) were compared. Results: The SSIM was 0.99 ± 9.15×10-5 and 0.99 ± 3.38×10-3 at 120 kVp and Sn140 kVp, respectively. The PSNR was 69.1 ± 1.29 and 66.5 ± 5.40 at 120 kVp and Sn140 kVp, respectively. Additionally, no significant differences were observed in the mean CT values of PA (p > 0.05) between the real and pseudo images. Conclusion: The proposed model accurately generated 120 kVp and Sn140 kVp CT images from 80 kVp CT images.
Send email to the article author
