Author: Debora de Souza Antonio, Romy Guthier, Konrad Pawel Nesteruk, Erno Sajo, William Paul Segars, Gregory C. Sharp, Atchar Sudhyadhom, Hengyong Yu 👨🔬
Affiliation: Massachusetts General Hospital, Carl E. Ravin Advanced Imaging Laboratories and Center for Virtual Imaging Trials, Duke University Medical Center, Brigham and Women’s Hospital and Dana Farber Cancer Institute, Harvard Medical School,, Massachusetts General Hospital and Harvard Medical School, University of Massachusetts Lowell 🌍
Purpose: To develop an open-access toolkit for rapidly generating simultaneously realistic CT scans and low-field MR images of the abdominal region, based on patient data, while employing an XCAT phantom geometry.
Methods: We produce two sets of image files of the XCAT phantom: one of realistic 120kVp CT scans, and another is realistic 0.35T MR images. HU and MRI intensities were derived from patient images obtained from a 0.35T MRIdian Linac using a bSSIP sequence, and a Siemens SOMATOM Confidence simulator. To make the simulated images more realistic, we include random HU and MRI intensity variations following a normal distribution with the mean value and standard deviation matching patient images. To validate the tool, histograms of ROI within organs of the synthetic and patient images were evaluated using the Kolmogorov-Smirnov distance. The mean values and standard deviations were compared for ROIs within three distinct organs. A normal distribution was fitted to the histograms to assess the noise approximation.
Results: CT and MR data sets were co-registered to reproduce the desired XCAT phantom geometry. The analysis of ROI within sampled organs using the generated CT and MR images vs the real patient images show good agreement. Noise within patient images did not follow a normal distribution, however normal distributions fit to the histograms, and show it is a reasonable approximation for most research purposes.
Conclusion: We developed and validated a tool to generate a pair of realistic synthetic CT and low-field MR images of the user-defined XCAT phantom anatomy. The developed toolkit allows generating simultaneously both co-registered image modalities, which is often required when “ground truth” is needed for multimodality studies. Our tool does not require substantial computational power or advanced programming skills.