Synthetic CT for MR-Only Prostate Radiotherapy Planning

CSC Lead: Anil

Patients prescribed radiotherapy to the prostate have a CT scan which provides anatomical information used to target the RT, and electron density info required for dose calculation in the treatment planning system (TPS). The patient’s RT plan is generated using the CT scan and the plan is sent to the treatment machine. Prostate RT patients experience toxicity side-effects due to irradiation of healthy tissues including rectum, bladder, urethra and penile bulb. Doses to these tissues are higher than they could be due to uncertainties in defining the prostate treatment volume on CT, where studies show that delineation of the prostate on MR results in 25-40% reduction in prostate volume [2-5] with more consistency between operators [6-9].

An in-house study at GSTT has shown that rectal and bladder dose can be reduced, and this should relate to a reduction in toxicity risk. The improved soft-tissue contrast of MRI can be used to better define targets and additional healthy tissue [10]. Ideally an MR-only pathway is used, where the MRI is used for anatomical delineation and for dose calculation. A benefit of MR-only RT is the removal of registration uncertainties [12] present in a combined CT + MRI pathway. MRI data has no inherent electron density information so for an MR-only pathway to be feasible a SynCT must be made for the TPS dose calculation.

Funding from the National Institute for Health Research and Social Care (NIHR) (RP-2-16-07-001) has supported Christopher Thomas during his PhD project which involved generating synthetic CT images from MRI images for prostate radiotherapy patients at Guy’s and St Thomas’s NHS FT. An output of the project is a UNET model that has been trained on registered CT and MRI image sets from 24 patients participating in a clinical trial evaluating MR-Only radiotherapy treatment planning (NCT03238170). This trained model is to be deployed into an MR-only clinical pathway via dicomserver.

Clinical lead: Christopher Thomas

Project Plan
1. Meeting of all persons involved to determine AI specifications.

2. Setting technical and system requirements for AI model.

3. Dataset curation (retrospective).

4. Model training

5. Model testing

6. Implementation

7. Audit

References
[1] CRUK prostate Cancer Statistics
[2] Rasch et al. 1997
[3] Sannazzari et al. 2014
[4] Roach et al. 1996
[5] Kagawa et al. 1997
[6] Debois et al. 1999
[7] Parker et al. 2003
[8] Pathmanathan et al. 2019
[9] Usmani et al. 2011
[10] Vainshtein et al. 2021
[11] Menard et al. 2018
[12] Nyholm et al. 2009
[13] Johnstone et al. 2018
[14] Thomas et al. 2020
[15] Bird et al. 2019
[16] Wyatt et al. 2019