Mixed effect model confirms increased risk of image changes with increasing linear energy transfer in proton therapy of gliomas.

Abstract

Background and purpose: Radiation induced image changes (IC) on MRI have been observed after proton therapy for brain tumours. This study aims to create predictive models, with and without taking into account patient variation, based on dose, linear energy transfer (LET) and periventricular zone (PVZ) in a national cohort of patients with glioma treated with pencil beam scanning (PBS).

Materials and methods: A cohort of 87 consecutive patients with oligodendroglioma or astrocytoma (WHO grade 2-4) treated with PBS from January 2019 to December 2021 was included. All patients were treated with three to four beams. Monte Carlo calculations of dose and LET were performed for all treatment plans. Lesion weighted as well as mixed effect logistic regression models were developed to predict IC in a voxel.

Results: 12 patients (14 %) developed ICs on the follow-up MR-scans. Mixed effect modelling accounting for interpatient variation was justified by the non-negligible inter class correlation coefficient (ICC = 0.33). The two approaches identified similar model features and marginal improvement in model performance was found, when increasing model parameters from two (AUC = 0.92/0.94) to three (AUC = 0.93/0.95) parameters. Univariate analysis showed that patients treated with narrow beam configurations had an increased incidence of IC (p = 0.01).

Conclusion: 14% of patients developed IC following PT. Lesion-weighted and mixed effect models resulted in similar model performance confirming increased risk of IC with increasing LET. The beam arrangement seems to influence the risk of IC and needs further investigation.