Multi-parametric and multi-regional histogram analysis of MRI: modality integration reveals imaging phenotypes of glioblastoma

Introduction Glioblastoma is characterized by its remarkable heterogeneity and dismal prognosis. Histogram analysis of quantitative magnetic resonance imaging (MRI) is an important in vivo method to study intratumoral heterogeneity. With large amounts of histogram features generated, integrating these modalities effectively for clinical decision remains a challenge. Methods A total of 80 patients with supratentorial primary glioblastoma were recruited. All patients received surgery and standard regimen of temozolomide chemoradiotherapy. Diagnosis was confirmed by pathology. Anatomical T2-weighted, T1-weighted post-contrast and FLAIR images, as well as dynamic susceptibility contrast (DSC), diffusion tensor imaging (DTI) and chemical shift imaging were acquired preoperatively using a 3T MRI scanner. DTI-p, DTI-q, relative cerebral blood volume (rCBV), mean transit time (MTT) and relative cerebral blood flow (rCBF) maps were generated. Contrast-enhancing (CE) and non-enhancing (NE) regions of interest were manually delineated. Voxel intensity histograms were constructed from the CE and NE regions independently. Patient clustering was performed by the Multi-View Biological Data Analysis (MVDA) approach. Kaplan-Meier and Cox proportional hazards regression analyses were performed to evaluate the relevance of the patient clustering to survival. The histogram features selected from MVDA approach were evaluated using receiver operator characteristics (ROC) curve analysis. The metabolic signatures of the patient clusters were analyzed by multivoxel MR spectroscopy (MRS). Results The MVDA approach yielded two final patient clusters, consisting of 53 and 27 patients respectively. The two patient subgroups showed significance for overall survival (p = 0.007, HR = 0.32) and progression-free survival (p Discussion This study demonstrated that integrating multi-parametric and multi-regional MRI histogram features may help to stratify patients. The histogram features selected from the proposed approach may be used as potential imaging markers in personalized treatment strategy and response determination.The research was supported by the National Institute for Health Research (NIHR) Brain Injury MedTech Cooperative based at Cambridge University Hospitals NHS Foundation Trust and University of Cambridge. The views expressed are those of the author(s) and not necessarily those of the NHS, the NIHR, or the Department of Health and Social Care (SJP, project reference NIHR/CS/009/011); CRUK core grant C14303/A17197 and A19274 (FM lab); Cambridge Trust and China Scholarship Council (CL & SW); the Chang Gung Medical Foundation and Chang Gung Memorial Hospital, Keelung, Taiwan (JLY); the Commonwealth Scholarship Commission and Cambridge Commonwealth Trust (NRB); CRUK & EPSRC Cancer Imaging Centre in Cambridge & Manchester (FM & TT, grant C197/A16465); and NIHR Cambridge Biomedical Research Centre (TM & SJP)