Click here to

Session: Machine Intelligence in Image Processing and Motion Correction I [Return to Session]

Synthesizing Iodine Map From Non-Contrast Enhanced CT Via Deep-Learning Network

H Xie1, Y Lei1, T Wang1, J Roper1, B B Ghavidel1, M McDonald1, D S Yu1, X Tang2, J D Bradley1, T Liu1, X Yang1*, (1) Department of Radiation Oncology and Winship Cancer Institute, (2) Department of Radiology and Imaging Sciences and Winship Cancer Institute, Emory University School Of Medicine, Atlanta, GA


TH-E-BRC-6 (Thursday, 7/14/2022) 1:00 PM - 2:00 PM [Eastern Time (GMT-4)]

Ballroom C

Purpose: Iodine maps generated with contrast enhanced dual-energy CT (CE-DECT) are valuable for radiation oncologists to contour target and organs-at-risk (OARs) in radiotherapy planning.However, the iodine intensity in the images could cause problems for dose calculation, especially during proton therapy treatment planning. The purpose of this work is to propose a deep-learning based method for synthesizing iodine maps from non-contrast enhanced CT images.

Methods: Images of 60 head and neck cancer patients were retrospectively investigated. During the simulation, the non-contrast enhanced planning CT (NonCECT) scan was followed by a CE-DECT scan. Pre-registered NonCECT images and iodine maps, which were generated from the CE-DECT scan, were treated as the input and ground truth (gtIodine) images. A deep-learning-based method, namely Cycle-Net, was crafted to generate iodine (synIodine) maps from the NonCECT images. The Cycle-Net was trained by several supervision mechanisms that optimize the learnable parameters to generate synthetic iodine maps. Cycle-Net is composed of two generators and two discriminators. The generators are used to build the forward and backward mapping between the NonCECT and iodine images. The discriminators are used to assess the quality of the synIodine images.

Results: The efficacy and accuracy of the proposed Cycle-Net was evaluated by several quantitative metrics comparing the synIodine images with the gtIodine images: Mean absolute error (19.18±8.67 HU), normalized cross correlation coefficient (0.97±0.01), structural similarity index (0.88±0.03), and peak signal-to-noise ratio (24.90±2.13 dB).

Conclusion: A deep-learning based method is proposed and validated for iodine map synthesis from NonCECT images. The proposed approach could be promising to eliminate the CE-DECT scan during specialized radiotherapy simulation, which could reduce imaging dose, labor and financial costs, as well as enhance patient comfort, while retaining the ability to provide synthetic iodine maps to the physicians. This technique may be broadly important for patients not eligible for iodine contrast.


Not Applicable / None Entered.


Not Applicable / None Entered.

Contact Email