Author: Jared Baggett, Wesley E. Bolch, Lukas M. Carter, Robert Joseph Dawson, Laura Dinwiddie, Gunjan Kayal, Adam Leon Kesner, Harry Marquis, Juan Camilo Ocampo Ramos, Stefan Wehmeier 👨🔬
Affiliation: Department of Medical Physics, Memorial Sloan Kettering Cancer Center, University of Florida, MD Anderson 🌍
Purpose: Accurate organ dose estimation in computed tomography (CT) has gained attention due to the increasing global utilization of CT imaging. MIRDct, a specialized CT dosimetry software, has recently been developed and is freely available on MIRDsoft.org. This work assesses its performance across standard clinical protocols and compares it with alternative tools.
Methods: Organ doses (in-field, partially-in-field, out-of-field) were evaluated across routine adult CT protocols (abdomen/pelvis, head, chest, whole-body) for male and female phantoms using scanners from GE, Philips, Siemens, and Toshiba. Simulations covered 28 cases with varying tube energies (100–140 kVp), collimation widths (10–40 mm), and CTDI phantom (head, body). CTDIvol, dose coefficients (organ doses normalized to CTDIvol), and effective doses from MIRDct were compared with NCICT3.0 and VirtualDose. Median relative differences (MdRD%) were calculated using MIRDct as the reference.
Results: CTDIvol comparisons across vendors showed variability ranging from 0 to 3-fold, with MIRDct aligning closer to NCICT than VirtualDose. For dose coefficients, the MdRD for in-field organs was 26% and 17% relative to NCICT and VirtualDose, respectively, rising to 46% and 76% for out-of-field organs. VirtualDose exhibited the highest variability, with differences for partially-in-field and out-of-field organs exceeding 100%. Effective dose comparisons showed greater variation between MIRDct and NCICT than VirtualDose for head (-25% vs. -42%) and whole-body (23% vs. 30%) protocols, but not for chest (40% vs. -6%) and AP (37% vs. -11%).
Conclusion: We observed significant variability in CTDIvol used to normalize doses across different software. Dose coefficients from MIRDct demonstrated reasonable agreement with NCICT and VirtualDose typically within 30% for in-field organs, while larger discrepancies were noted for out-of-field organs. These differences can be attributed to variations in scan range, phantom geometries, and reference scanners used in Monte Carlo simulations. These findings highlight the inherent variations among CT dose modeling software.