A multi‐institutional analysis of a general pelvis continuous Hounsfield unit synthetic CT software for radiotherapy

Abstract Purpose To validate a synthetic computed tomography (sCT) software with continuous HUs and large field‐of‐view (FOV) coverage for magnetic resonance imaging (MRI)‐only workflow of general pelvis anatomy in radiotherapy (RT). Methods An sCT software for general pelvis anatomy (prostate, rectum, and female pelvis) has been developed by Philips Healthcare and includes continuous HUs assignment along with large FOV coverage. General pelvis sCTs were generated using a two‐stack T1‐weighted mDixon fast‐field echo (FFE) sequence with a superior‐inferior coverage of 36 cm. Seventy‐seven prostate, 43 rectum, and 27 gynecological cases were scanned by three different institutions. mDixon image quality and sCTs were evaluated for soft tissue contrast by using a confidence level scale from 1 to 5 for bladder, prostate/rectum interface, mesorectum, and fiducial maker visibility. Dosimetric comparison was performed by recalculating the RT plans on the sCT after rigid registration. For 12 randomly selected cases, the mean absolute error (MAE) between sCT and CT was calculated to evaluate HU similarity, and the Pearson correlation coefficients (PCC) between the CT‐ and sCT‐generated digitally reconstructed radiographs (DRRs) were obtained for quantitative comparison. To examine geometric accuracy of sCT as a reference for cone beam CT (CBCT), the difference between bone‐based alignment of CBCT to CT and CBCT to sCT was obtained for 19 online‐acquired CBCTs from three patients. Results Two‐stack mDixon scans with large FOV did not show any image inhomogeneity or fat‐water swap artifact. Fiducials, Foley catheter, and even rectal spacer were visible as dark signal on the sCT. Average visibility confidence level (average ± standard deviation) on the sCT was 5.0 ± 0.0, 4.6 ± 0.5, 3.8 ± 0.4, and 4.0 ± 1.1 for bladder, prostate/rectum interface, mesorectum and fiducial markers. Dosimetric accuracy showed on average < 1% difference with the CT‐based plans for target and normal structures. The MAE of bone and soft tissue between the sCT and CT are 120.9 ± 15.4 HU, 33.4 ± 4.1 HU, respectively. Average PCC of all evaluated DRR pairs was 0.975. The average offset between CT and sCT as reference was (LR, AP, SI) = (0.19 ± 0.35, 0.14 ± 0.60, 0.44 ± 0.54) mm. Conclusions The continuous HU sCT software‐generated realistic sCTs and DRRs to enable MRI‐only planning for general pelvis anatomy.


Conclusions:
The continuous HU sCT software-generated realistic sCTs and DRRs to enable MRI-only planning for general pelvis anatomy.

K E Y W O R D S
mDixon, MR-only, multi-institutional, pelvis, radiotherapy, synthetic CT

| INTRODUCTION
Magnetic resonance imaging (MRI) has been an integral part of the radiotherapy (RT) process for more than a decade due to its excellent soft tissue contrast. Multiple studies have shown the superiority of MRI for target and normal tissue segmentation in external beam RT by demonstrating reduced interobserver variability in contours compared with those obtained from computed tomography (CT). [1][2][3] The current RT simulation process relies on target and organs-at-risk (OAR) segmentation on MRI followed by transfer of contours to CT via image registration. Although incorporating MRI decreases over-segmentation of structures as compared with CT-based segmentation, a combined CT-MRI method is challenging due to errors introduced by mis-registration of the image sets and the changes to the shape and location of the soft tissues, for example, bladder, rectum, and seminal vesicles that are inherent when acquiring multiple image sets. [4][5][6] Because of the challenges in target delineation, registration uncertainties, and changes in anatomy due to temporal variations, a workflow in which MRI is the primary and sole imaging modality is highly preferable to a combined CT and MRI workflow.
An important component of MRI-only workflow is generation of synthetic CTs (sCT). An sCT software for general pelvis anatomy (prostate, rectum, and female pelvis) has been developed by Philips Healthcare and includes continuous Hounsfield units (HU) assignment along with large field-of-view (FOV) coverage. The method is an extension of the earlier MRCAT (magnetic resonance for calculating attenuation) prostate software package that is currently implemented clinically at various institutions. 7-10 MRCAT prostate uses a single 3D mDIXON XD fast-field echo (FFE) scan to generate sCTs. A constrained shape model is used to estimate body contour as well as segment bone structures. Bulk densities are assigned to five different tissue types (air, fat, soft tissue, spongy bone, and cortical bones). MRCAT is limited to a superior-inferior extent of 30 cm and up to L4 vertebrae only. MRCAT prostate has recently been modified using a two-stack mDIXON sequence and developed to generate sCTs using continuous HU generation as well as for general pelvis anatomy. For many high-risk prostate, gynecological, and rectum cases nodal volumes are treated which can extent up to L1-L3.
Scanning larger volume in superior-inferior direction is challenging due to the concerns for geometrical distortions as well as scan homogeneity. The goal of this study was to assess the image quality, dosimetric and geometric accuracy of an sCT software with continuous HUs and large FOV coverage for MRI-only workflow of a general pelvis anatomy.

2.C | Data analysis
Image quality of the mDIXON source images and sCTs from general pelvis cases from all three institutions were evaluated for soft tissue contrast by using a confidence level scale from 1 to 5 for bladder, prostate/rectum interface, mesorectum, and fiducial maker visibility.
To evaluate the accuracy of the sCT for patient treatment planning, the treatment plan and structure set from the original planning online-acquired CBCTs from three patients.

3.A | mDIXON MRI and synthetic CT image quality
Large FOV acquisition is challenging in terms of geometric accuracy in the SI direction. Two-stack mDixon acquisition mitigated that. Large    deviation) offset between CT and sCT as reference was (LR, AP, SI) = (0.19 AE 0.35, 0.14 AE 0.60, 0.44 AE 0.54) mm. A box plot of the resultant offset in all three directions is shown in Fig. 6(b).

MAE of bone, soft tissue, and total body of the sCT-BD
approach compared with the sCT-cHU approach is shown in Table 3. sCT-cHU has lower MAE values than that of sCT-BD for all compared tissue volumes as expected, particularly significant decrease of up to 40.3 HU was shown for the MAE in bone. A HU profile comparison of the two methods and the CT is also shown in Fig. 2 for an example case, where a closer agreement between the sCT-cHU to CT can be seen.

| DISCUSSION
Successful clinical implementation of MRI-only treatment planning have been reported from multiple institutions across the world but remains primarily in prostate cancer patients. 7,8,11,12 Wider utilization of MRI-only treatment planning to maximize the benefit from the superior soft tissue contrast in MRI remains to be seen in other T A B L E 2 Tissue-specific mean absolute error (MAE) between synthetic CT and CT for bone, soft tissue, and body (CT: computed tomography; HU: Hounsfield unit, Pt: patient).  pelvis disease sites such as rectal and gynecological cancers. In rectal cancer, MRI is the most accurate tool for local cancer staging 13,14 and a powerful method to determine best course of treatment. [15][16][17] In gynecological cancers, consensus remains that MRI is significantly more reliable than CT for the delineation of gross tumor volume, adjacent uterine tissue, and superior/inferior bladder extent. 18,19 Studies have also demonstrated that MRI-assisted dose escalation in gynecological brachytherapy allowed for 10-20% overall survival gains while reducing urinary and gastrointestinal late morbidity. [20][21][22] In a multi-institutional study setting, we investigated the soft tis-  This study serves as a complement and extension to Kemp-

| CONCLUSION S
MRCAT general pelvis with continuous HU generated realistic sCTs and DRRs to enable MRI-only planning for general pelvis anatomy.
Two-stack acquisition enabled geometrically accurate MRI as well as sCT images and allowed anatomic coverage up to L1-L3 vertebrae to enable treatment to superior nodal volumes. The extension to T A B L E 3 Mean absolute error (MAE) comparison for continuous HU synthetic CT (sCT-cHU) and bulk density synthetic CT (sCT-BD) to CT for two example cases (CT: computed tomography; HU: Hounsfield unit).