Girolamo Crisi1, Silvano Filice2. 1. Neuroradiology Unit, Azienda Ospedaliero-Universitaria of Parma, Parma, Italy. 2. Medical Physics Unit, Azienda Ospedaliero-Universitaria of Parma, Parma, Italy.
Abstract
BACKGROUND AND PURPOSE: This study aims to investigate whether radiomic quantitative image features (IFs) from perfusion dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI) retain sufficient strength to predict O6-methylguanine-DNA methyltransferase promoter methylation (MGMT_pm) in newly diagnosed glioblastoma (GB) patients. METHODS: We retrospectively reviewed the perfusion DSC-MRI of 59 patients with GB. Patients were classified into three groups: (1) unmethylated if MGMT_pm ≤ 9% (UM); (2) intermediate-methylated if MGMT_pm ranged between 10% and 29% (IM); (3) methylated if MGMT_pm ≥ 30% (M). A total of 92 quantitative IFs were obtained from relative cerebral blood volume and relative cerebral blood flow maps. The Mann-Whitney U-test was applied to assess whether there were statistical differences in IFs between patient groups. Those IFs showing significant difference between two patient groups were termed relevant IFs (rIFs). rIFs were uploaded to a machine learning model to predict the MGMT_pm. RESULTS: No rIFs were found between UM and IM groups. Fourteen rIFs were found among UM-M, IM-M, and (UM + IM)-M groups. We built a multilayer perceptron deep learning model that classified patients as belonging to UM + IM and M group. The model performed well with 75% sensitivity, 85% specificity, and an area under the receiver-operating curve of .84. CONCLUSION: rIFs from perfusion DSC-MRI are potential biomarkers in GBs with a ≥30% MGMT_pm. Otherwise, unmethylated and intermediate-methylated GBs lack of rIFs. Five of 14 rIFs show sufficient strength to build an accurate prediction model of MGMT_pm.
BACKGROUND AND PURPOSE: This study aims to investigate whether radiomic quantitative image features (IFs) from perfusion dynamic susceptibility contrast magnetic resonance imaging (DSC-MRI) retain sufficient strength to predict O6-methylguanine-DNA methyltransferase promoter methylation (MGMT_pm) in newly diagnosed glioblastoma (GB) patients. METHODS: We retrospectively reviewed the perfusion DSC-MRI of 59 patients with GB. Patients were classified into three groups: (1) unmethylated if MGMT_pm ≤ 9% (UM); (2) intermediate-methylated if MGMT_pm ranged between 10% and 29% (IM); (3) methylated if MGMT_pm ≥ 30% (M). A total of 92 quantitative IFs were obtained from relative cerebral blood volume and relative cerebral blood flow maps. The Mann-Whitney U-test was applied to assess whether there were statistical differences in IFs between patient groups. Those IFs showing significant difference between two patient groups were termed relevant IFs (rIFs). rIFs were uploaded to a machine learning model to predict the MGMT_pm. RESULTS: No rIFs were found between UM and IM groups. Fourteen rIFs were found among UM-M, IM-M, and (UM + IM)-M groups. We built a multilayer perceptron deep learning model that classified patients as belonging to UM + IM and M group. The model performed well with 75% sensitivity, 85% specificity, and an area under the receiver-operating curve of .84. CONCLUSION: rIFs from perfusion DSC-MRI are potential biomarkers in GBs with a ≥30% MGMT_pm. Otherwise, unmethylated and intermediate-methylated GBs lack of rIFs. Five of 14 rIFs show sufficient strength to build an accurate prediction model of MGMT_pm.
Authors: Vittorio Stumpo; Lelio Guida; Jacopo Bellomo; Christiaan Hendrik Bas Van Niftrik; Martina Sebök; Moncef Berhouma; Andrea Bink; Michael Weller; Zsolt Kulcsar; Luca Regli; Jorn Fierstra Journal: Cancers (Basel) Date: 2022-03-05 Impact factor: 6.639