Résumé: O-6-methylguanine-DNA methyltransferase (MGMT) is one of the most salient gene promoters that correlates with the effectiveness of standard therapy for patients suffering from glioblastoma (GBM). Non-invasive estimation of MGMT and overall survival (OS) in GBM patients could provide a particular direction to neuro-oncologists and surgeons for precise treatment and surgical planning. This study investigated hybrid radiomics signatures (HRS) for the prediction of (i) MGMT status (methylated/unmethylated) and (ii) OS (short survivors 12 months and long survivors?>??=?12 months) using both conventional and deep radiomic features derived from multi-parametric MRI (mp-MRI). Further, for the OS, Kaplan–Meier analysis was carried out to analyze the difference between two groups of survivors. Additionally, Cox-PH modeling was adapted to investigate the impact of clinical observation on OS. Two cohorts of 555 and 209 GBM patients have been used to analyze HRS for MGMT and OS, respectively. (i) For MGMT status prediction employing conventional machine learning radiomics features along with deep learning features using VGG16 and VGG19, the HRS obtained an AUC of 0.76 (95% CI: 0.70–0.80). (ii) For OS prediction employing the log-rank test, the conventional radiomic signature showed an AUC of 0.78 (95% CI: 0.75–0.83) with a p-value?<?0.001. Similarly, in assessing the impact of patient age on OS, the concordance index was 0.68 (95% CI 0.6–0.72). The proposed study concludes with the diagnostics remark of efficient HRS for MGMT prediction and conventional radiomics for OS prediction.

Langue: Anglais