Divergent white matter changes in patients with nasopharyngeal carcinoma post-radiotherapy with different outcomes: a potential biomarker for prediction of radiation necrosis

Abstract

Objectives: To investigate the effects of standard radiotherapy on temporal white matter (WM) and its relationship with radiation necrosis (RN) in patients with nasopharyngeal carcinoma (NPC), and to determine the predictive value of WM volume alterations at the early stage for RN occurrence at the late-delay stage.

Methods: Seventy-four treatment-naive NPC patients treated with standard radiotherapy were longitudinally followed up for 36 months. Structural MRIs were collected at multiple time points during the first year post-radiotherapy. Longitudinal structural images were processed using FreeSurfer. Linear mixed models were used to delineate divergent trajectories of temporal WM changes between patients who developed RN and who did not. Four machine learning methods were used to construct predictive models for RN with temporal WM volume alterations at early-stage.

Results: The superior temporal gyrus (STG) had divergent atrophy trajectories in NPC patients with different outcomes (RN vs. NRN) post-radiotherapy. Patients with RN showed more rapid atrophy than those with NRN. A predictive model constructed with temporal WM volume alterations at early-stage post-radiotherapy had good performance for RN; the areas under the curve (AUC) were 0.879 and 0.806 at 1-3 months and 6 months post-radiotherapy, respectively. Moreover, the predictive model constructed with absolute temporal volume at 1-3 months post-radiotherapy also presented good performance; the AUC was 0.842, which was verified by another independent dataset (AUC = 0.773).

Conclusions: NPC patients with RN had more sharp atrophy in the STG than those with NRN. Temporal WM volume at early-stage post-radiotherapy may serve as an in vivo biomarker to identify and predict RN occurrence.

Key points: • The STG had divergent atrophy trajectories in NPC patients with different outcomes (RN vs. NRN) post-radiotherapy. • Although both groups exhibited time-dependent atrophy in the STG, the patients with RN showed a more rapid volume decrease than those with NRN. • Temporal WM volume alteration (or absolute volume) at the early stage could predict RN occurrence at the late-delay stage after radiotherapy.

Keywords: Machine learning; Nasopharyngeal carcinoma; Radiation-induced temporal lobe necrosis; Radiotherapy; White matter.