Iodine maps can be obtained from contrast enhanced dual-energy compute tomography (DECT) scans to emphasize iodine contrast agent uptake in cancer patients’ tissues, which benefits radiation oncologists in the treatment planning process. However, DECT scanners are not widely equipped among the radiation therapy centers. Furthermore, certain patients, i.e., either with iodine allergies or renal dysfunction, are not suitable for iodine contrast DECT scans. The purpose of this work is to generate synthetic iodine maps based on non-contrast single-energy CT (SECT) images via deep learning (DL) method. 130 head-and-neck patients’ images were retrospectively investigated in this work. All patients were scanned with non-contrast SECT and contrast DECT protocols. The ground truth iodine maps were generated from contrast DECT scans using vender software. A denoising diffusion probabilistic model (DDPM) was implemented to generate synthetic iodine maps. The training and application datasets were kept strictly separate, containing data from 100 and 8 patients respectively. A CycleGAN was implemented as a reference method to assess the proposed DDPM method. The accuracy of the proposed DDPM was evaluated using three quantitative metrics: Mean absolute error (MAE) (19.31±3.38 HU), structural similarity index (0.79±0.13) and peak signal-to-noise ratio (22.25±4.23dB) respectively. Compared to the reference method, the proposed method demonstrated superior performance, which was further corroborated by paired two-tailed t-tests, across these metrics. To our best knowledge, this work is the first of its kind to demonstrate the capability to provide synthetic iodine maps based on SECT via DDPM method.
|