Early Differentiation of Pseudoprogression for Recurrent Glioblastoma in STERIMGLI Trial Using Multimodal MRI Radiomics Signature

Abstract

<h3>Purpose/Objective(s)</h3> Glioblastoma multiforme (GBM) is the most common and aggressive primary brain tumor in adults. Local relapse of GBM is constant and leads to death. STERIMGLI trial (NCT02866747) evaluates the efficacy of combined stereotactic radiotherapy with Durvalumab as salvage treatment for patients with recurrent GBM. However, immunotherapy leads to morphological pseudoprogression (PP) images. Interestingly, patients in phase I who presented a PP had a prolonged local control. Up to date, no early imaging biomarkers are able to predict which patients will present a PP and potentially respond to the combined treatment. The study aims at identifying early predictive imaging biomarkers to characterize PP from true progression. <h3>Materials/Methods</h3> 26 patients from STERIMGLI phase I/IIa with longitudinal multi-parametric MRI follow-up (morphological MRI modalities, diffusion, perfusion) were included. BraTS Toolkit was used for image preprocessing and first automatic segmentation of tumor subregions (<i>Piram et al. EANO 2021</i>) named here after labels enhanced tumor (ET) and flair edema (ED). Segmentations were reviewed by a radiation oncologist and neuroradiologist, and MRIs were classified as PP or true progression (TP). First order radiomic features were extracted with open-source software in FLAIR, ADC map, rCBV map for ED label and T1, T1c, ADC, rCBV for ET label. Extracted features were compared at baseline, first event and variation (∆) between baseline and first event, in TP and PP groups, for tumor (ET, ED) and contralateral healthy tissue (control ET, control ED). <h3>Results</h3> On baseline FLAIR, kurtosis in ED label predicts PP (K< 2.52) and TP (K> 2.52) outcome (p= 0.0047). Sensitivity and specificity were respectively 76% and 89% in our study group. In ET label, ∆volume (p=0.0055), ∆entropy on 3 modalities T1 (p=0.0142), T1c (p= 0.0253) and ADC map (p= 0.001) showed significant differences between TP and PP groups on univariate analysis, whereas no difference was observed in healthy tissue. A combined score of those 4 features, was able to differentiate patients in two groups: low risk of TP versus high risk of TP and discriminate between TP and PP with 87% sensitivity and 88% specificity. Odd ratio between the 2 groups is 19.50 [2.2; 173.5]. <h3>Conclusion</h3> Radiomics analysis was able to identify a predictive feature of PP on baseline FLAIR MRI and a diagnostic combined score was able to discriminate between TP and PP, from variation between first significant volume increase and baseline MRI, in patients treated with combined stereotactic radiotherapy with Durvalumab for recurrent GBM. However, these results have to be confirmed on a larger cohort (on-going STERIMGLI phase IIb will provide a validation cohort)