Abstract:
Malignant glioma is a type of frequent and lethal cancer the brain. Recent World Health Organization (WHO) criteria has included genetic mutations in glioma classification. One of these mutations, isocitrate dehydrogenase (IDH) is common in grades II and III gliomas, and has been related to metabolism of the cancer tissue. IDH mutant gliomas have better prognosis than IDH wild type ones. As a result of this mutation, an onco-metabolite 2-HydroxyGlutarate (2HG) accumulates in tumor tissue. Detection of IDH mutation before surgical procedure could play an important role in treatment planning. Magnetic resonance spectroscopy (MRS) is a noninvasive technique that could be used to provide IDH mutation information. In this study, first, a 3D printed MRS phantom was designed and produced to analyze spatial distribution performances of MRS sequences. Then, 82 glioma patients, whose IDH status have been determined by immunohistochemistry, have been included. Short echo time Point ResolvedSpectroscopy(PRESS)andMescher-GarwoodPRESS(MEGA-PRESS)MRS sequences were acquired on a 3T Siemens MRI scanner. Metabolite concentrations have been estimated with LCModel spectal fitting program using corresponding basis sets. Machine learning models have been developed to determine IDH mutation using metabolite concentrations as features. Our results indicated that a decision tree model using features from short TE PRESS profile could detect IDH mutation with 75% accuracy, while maximum accuracy attainable with MEGA-PRESS was 68%. The MRS phantom that was produced as a part of this study could be used as a validation tool for new MRS sequences. Future studies will aim to detect other genetic alterations in gliomas on a larger patient cohort.|Keywords : glioma, IDH, magnetic resonance spectroscopy, machine learning.
