Copenhagen, Denmark
Onsite/Online

ESTRO 2022

Session Item

Saturday
May 07
16:55 - 17:55
Poster Station 1
07: Imaging & AI techniques
Stephanie Tanadini-Lang, Switzerland
Poster Discussion
Physics
Two-ways validation of the feasibility of AI-based synthetic CT for MR-only brain radiotherapy
Siti Masitho, Germany
PD-0320

Abstract

Two-ways validation of the feasibility of AI-based synthetic CT for MR-only brain radiotherapy
Authors:

Siti Masitho1, Juliane Szkitsak1, Johanna Grigo1, Florian Putz1, Rainer Fietkau1, Christoph Bert1

1University Hospital Erlangen, Radiation Oncology, Erlangen, Germany

Show Affiliations
Purpose or Objective

Recent advances in MRI techniques enable to perform the RT workflow entirely with only MR-imaging. For this MR-only workflow, certain MRI images can be converted to a synthetic CT (sCT). sCTs should provide similar HU values as planning CTs for dose calculation and sufficient DRRs for daily positioning. A new artificial intelligence (AI)-based sCT generation technique based on a single T1-VIBE DIXON sequence was evaluated w.r.t. dose calculation and daily positioning for the brain radiotherapy workflow.

Material and Methods

The T1-VIBE DIXON (in-phase and opp-phase, 1.5x1.5x1.5mm3) was acquired at the 1.5 T MRI (Magnetom Sola, Siemens Healthineers) for eight brain RT patients with various indications (including metastasis, glioma, and whole brain). Patients were scanned in RT setup, which utilizes an RT mask system, an RT flat table top, and a dedicated coil setup. Syngo.via VB60 (Siemens Healthineers) was used to convert the MRI images to sCTs. For each patient, an RT plan was created on the planning CT, which was then recalculated on the sCT (referred to as TPCT→sCT); and a second plan was created on the sCT, which was then recalculated on the planning CT (TPsCT→CT). To create comparable RT plans; the RT mask system, CT markers, and table were previously removed from the planning CT image. Mean absolute percentage deviation of D50 (∆D50[%]) and D0.01ccm (∆D0.01[%]) were evaluated for the target volumes and OARs, respectively. Using Exactrac v6.0.6 (Brainlab), DRRs from the sCT were fused with X-ray images recorded as part of the patient treatment. The difference between the calculated couch shift/rotation after fusion and the recorded couch shift using the planning CT was evaluated.

Results

∆D50[%] for PTV(TPCT→sCT)=0.28±0.27% and for PTV(TPsCT→CT)=0.25±0.25% were found (Fig.1). Meanwhile, ∆D50[%] for GTV(TPCT→sCT)=0.20±0.24% and for GTV(TPsCT→CT) =0.21±0.31% were obtained. Largest mean ∆D0.01[%] of OARs was found on the cochlea(L)(TPCT→sCT)=1.87±1.7% (max. ∆D0.01[%]=4.34%) and the optical nerve(R)(TPsCT→CT)=0.83±0.73%. (max. ∆D0.01[%]=2.17%). The overall mean ∆D0.01[%] of all OARs is 0.78±1.03% for TPCT→sCT and 0.43±0.49% for TPsCT→CT. For DRR – X-ray fusion, median of couch shifts difference in lat./long./vert. direction are -0.22±1.43mm/-0.56±2.68mm/-0.53±1.78mm and median of rotation difference in roll/yaw/pitch direction are 0.17±1.0°/0.07±0.49°/-0.29±0.95°. Overall, the mean ∆D50[%] and mean ∆D0.01[%] are <2% for PTV, GTV, and various OARs, and the median of shift/rotation difference during positioning are <1mm/1°.

Conclusion

The AI-based sCT seems to be comparable with the planning CT for dose calculation and daily positioning with stereoscopic X-ray imaging. The study is currently still ongoing, in which more sCT plans will be evaluated.