Click here to

Session: Machine Learning for Adaptive Radiotherapy [Return to Session]

Clinical Application of Synthetic CTs for Proton Therapy of Lung Cancer Patients

VT Taasti1*, D Hattu1, I Hadzic1, S Pai1, M Gooding2, J Sage2, D De Ruysscher1, A Traverso1, R Canters1. (1) Department of Radiation Oncology (Maastro), GROW School for Oncology, Maastricht University Medical Centre+, Maastricht, The Netherlands; (2) Mirada Medical Ltd, Oxford, United Kingdom

Presentations

MO-H345-IePD-F2-3 (Monday, 7/11/2022) 3:45 PM - 4:15 PM [Eastern Time (GMT-4)]

Exhibit Hall | Forum 2

Purpose: To evaluate the clinical applicability of synthetic-CTs (sCTs) created from cone-beam-CTs (CBCTs) for monitoring the need for adaptation in proton therapy.

Methods: Proton patients in our clinic receive weekly scheduled repeat-CTs (reCTs) to assess the need for plan adaptations. Additionally, daily CBCTs are acquired for patient positioning. To reduce reCTs, we evaluated two different models converting CBCTs to sCTs, based on deep-learning (DL; in-house) and deformable image registration (DIR; Mirada Medical). Similar to our clinical reCT-workflow, we performed robust dose evaluation on both the reCT and same-day sCT to assess if the plan was still clinically acceptable (clinical-target-volume (CTV) coverage). If the plan violated clinical constraints on the reCT but was acceptable on the sCT a false negative was scored. Additionally, the sCT quality was evaluated by comparing dose-volume-histogram parameters and average CT numbers (DeltaCTNs) for sCTs and reCTs with matched anatomy. Thirty-five lung cancer patients were included: In the first evaluation (clinical), fifteen adapted and five non-adapted; and in the second (quality), twenty non-adapted.

Results: For the clinical evaluation, the DL-sCT correctly assigned 15/20(75%) patients and DIR-sCT 16/20(80%); both had 4 false negatives. For the quality evaluation, the CTV D95% on average deviated 3.1 and 0.8 Gy for DL-sCT and DIR-sCT, respectively. The DL-sCT result was mainly affected by one outlier patient, excluding this patient 1.2 Gy deviations was seen. Heart mean dose deviated 0.4 Gy on average for both. DeltaCTNs for the heart were 18 and 7 HU for DL-sCT and DIR-sCT, respectively.

Conclusion: Similar results were obtained for both sCT models, and in ≥75% a correct adaptation flagging was obtained. Both models had shortcomings and advantages; DL-sCTs were less accurate for uncommon patient arm positionings but reproduced anatomical changes well, while unrealistic deformations of bones were seen for DIR-sCTs which though had high CT number stability.

Funding Support, Disclosures, and Conflict of Interest: This study is part of the project Making Radiotherapy sustainable with project number 10070012010002 of the Highly Specialised Care & Research programme (TZO programme) which is (partly) financed by the Dutch Organisation for Health Research and Development (ZonMw). MG & JS are Employees of Mirada Medical Ltd.

Keywords

Cone-beam CT, Image-guided Therapy, Protons

Taxonomy

TH- External Beam- Particle/high LET therapy: Proton therapy – adaptive therapy

Contact Email

Share: