CT-based synthetic contrast-enhanced virtual monoenergetic images generation using conditional denoising diffusion probabilistic model Conference Paper
| Authors: | Gao, Y.; Xie, H.; Chang, C. W.; Qiu, R.; Wang, T.; Roper, J.; Ghavidel, B.; Zhou, J.; Yang, X. |
| Title: | CT-based synthetic contrast-enhanced virtual monoenergetic images generation using conditional denoising diffusion probabilistic model |
| Conference Title: | Medical Imaging 2025: Imaging Informatics |
| Abstract: | The primary objective of this research is to leverage deep learning (DL) methodologies, specifically a Conditional-Denoising Diffusion Probabilistic Model (C-DDPM), to generate synthetic contrast-enhanced virtual monoenergetic images (CE-VMI) from non-contrast CT (NC-CT) images. CE-VMI is an advanced imaging technology generated by dual-energy CT (DECT) scanners, which can improve contrast resolution, reduce beam hardening, metal artifacts and noise level. However, DECT scanners are not commonly available in many radiation therapy centers. Additionally, the use of iodine contrast agents is not suitable for all patients, particularly those who have adverse reactions to contrast media, further limiting the application of CE-VMI. These constraints restrict the widespread adoption of CE-VMI, despite its potential benefits in enhancing imaging quality and diagnostic accuracy. To address those challenges, our proposed approach offers a valuable alternative method. In this research, a retrospective analysis was conducted on CT simulation images from 120 patients with head-and-neck cancer who underwent proton radiotherapy. For the purposes of this study, each patient underwent two types of scans: NC-CT, which is utilized for dose calculation, and CE-DECT, which facilitates the delineation of the target area and nearby organs at risk. These scans were performed in accordance with the prescribed protocols of radiation oncologists. 70 keV and 140 keV VMIs are generated with vendor software. The CE-VMIs were then meticulously aligned with the NC-CT images using both rigid and deformable registration techniques to ensure accurate superimposition. To synthesize CE-VMI from NC_CT data, the study employed a C-DDPM. 100 patients were randomly selected as training dataset, and the remaining 20 patients were selected as test dataset. For the synthetic CE-VMI, the mean absolute error (MAE) for 70 keV and 140 keV were 38.36±21.32 and 33.18±22.76 Hounsfield Units (HU), the structural similarity index (SSIM) for two energies stood at 0.92±0.05 and 0.93±0.07, and the peak signal-to-noise ratio (PSNR) were 28.69±9.32 decibels (dB) and 33.20±10.35 dB, respectively. © 2025 SPIE |
| Keywords: | radiotherapy; diagnostic radiography; contrast-enhanced; de-noising; image denoising; dual-energy ct; transillumination; probabilistic models; monoenergetic; synthetic imaging; sect; ddpm; photointerpretation; contrast imaging; virtual monoenergetic; intelligent agents |
| Journal Title | Progress in Biomedical Optics and Imaging - Proceedings of SPIE |
| Volume: | 13411 |
| Conference Dates: | 2025 Feb 16-21 |
| Conference Location: | San Deigo, CA |
| ISBN: | 1605-7422 |
| Publisher: | SPIE |
| Date Published: | 2025-01-01 |
| Start Page: | 134111D |
| Language: | English |
| DOI: | 10.1117/12.3048623 |
| PROVIDER: | scopus |
| DOI/URL: | |
| Notes: | Conference Paper (ISBN: 9781510686007) -- Source: Scopus |
Altmetric
Citation Impact
BMJ Impact Analytics
Related MSK Work