Prediction of MGMT promoter methylation in glioblastoma and grade 4 astrocytoma using fluid-suppressed chemical exchange saturation transfer MRI and machine learning-based segmentation.
Authors
Affiliations (13)
Affiliations (13)
- Division of Radiology, Department of Clinical Sciences, Lund University, Lund, Sweden.
- BioMedical Engineering and Imaging Institute, Icahn School of Medicine at Mount Sinai, New York, NY, United States.
- Division of Pathology, Department of Clinical Sciences, Lund University, Lund, Sweden.
- Advanced Systems, Magnetic Resonance, Siemens Healthineers AG, Erlangen, Germany.
- Innovation Lab, Olea Medical, La Ciotat, France.
- Department of Medical Imaging and Physiology, Skåne University Hospital, Lund, Sweden.
- Division of Neurology, Department of Clinical Sciences, Lund University, Lund, Sweden.
- Division of Neurosurgery and Kamprad Laboratory, Department of Clinical Sciences, Lund University, Lund, Sweden.
- F.M. Kirby Research Center for Functional Brain Imaging, Kennedy Krieger Institute, Baltimore, MD, United States.
- Department of Radiology, Johns Hopkins University School of Medicine, Baltimore, MD, United States.
- Department of Neurology, Johns Hopkins University School of Medicine, Baltimore, MD, United States.
- Department of Medical Radiation Physics, Lund University, Lund, Sweden.
- Lund University Bioimaging Center (LBIC), Lund University, Lund, Sweden.
Abstract
Treatment response in high grade gliomas (HGG) is influenced by O6-methylguanine-DNA methyltransferase promoter methylation (MGMTpm). Chemical exchange saturation transfer (CEST) magnetic resonance imaging (MRI) is a non-invasive approach for potential molecular tumor characterization, although previous studies have reported inconsistent results for MGMTpm. This study investigates fluid-suppressed (FS) CEST metrics and automated tumor segmentation in the preoperative prediction of MGMTpm in HGG. 3 T MRI including CEST imaging was performed in 44 patients with HGG (mean age 59 years, 16 female, 24 MGMTpm, 39 glioblastoma and five astrocytoma, grade 4). Contrast-enhancing tumor (ET), necrosis, and peritumoral edema were segmented manually and with two machine learning (ML) based models (DeepBraTumIA and Raidionics). Maximum, minimum and percentile-based metrics were extracted from FS amide proton transfer-weighted signal at 3.5 ppm (APTw) and FS CEST signal at 2.0 ppm (CEST@2ppm), normalized with contralateral normal-appearing white matter. The APTw/CEST@2ppm ratio was calculated. Statistical analysis was performed between MGMTpm and non-MGMTpm tumors using group comparisons, Spearman correlation, and receiver operating characteristics with area under curve (AUC), corrected for multiple comparisons. Significant differences were found in non-MGMTpm relative to MGMTpm tumors: higher 90th percentile and max APTw in necrosis across all segmentation methods; higher 90th percentile APTw, max APTw and max CEST@2ppm in ET with the ML-based models; higher max CEST@2ppm and max ratio in necrosis with Raidionics. The findings with APTw and CEST@2ppm remained consistent in glioblastoma patients and were further supported by significant inverse correlations between the CEST metrics and percentage of MGMTpm. The best-performing parameters for predicting MGMTpm status were the max APTw in ET and necrosis (AUC 0.76-0.82) and max CEST@2ppm in ET and necrosis (AUC 0.67-0.85), achieving higher sensitivity (69-100%) compared to specificity (59-82%). MGMTpm status in HGG may be predicted with fluid-suppressed CEST MRI. Using publicly available ML-based segmentation tools highlights a potential reproducible workflow in the clinical setting. Future multicenter studies with optimized acquisition protocols and independent, multimodal validation are warranted.