ONCOhabitats: A system for glioblastoma heterogeneity assessment through MRI

Abstract

[EN] Background: Neuroimaging analysis is currently crucial for an early assessment of glioblastoma, to help improving treatment and tumor follow-up. To this end, multiple functional and morphological MRI sequences are usually employed, requiring the development of automated tools capable to extract the relevant information from these sources. In this work we present ONCOhabitats (https://www.oncohabitats.upv.es): an online open access system for glioblastoma analysis based on MRI data.

Methods: ONCOhabitats provides two main services for untreated glioblastomas: (1) malignant tissue segmentation, and (2) vascular heterogeneity assessment of the tumor. The segmentation service implements a deep patch-wise 3D Convolutional Neural Network with residual connections. The vascular heterogeneity assessment service implements the Hemodynamic Tissue Signature (HTS) method patented in P201431289, which aims to identify habitats within the tumor with early prognostic capabilities.

Results: The segmentation service was validated against the BRATS 2017 reference dataset, showing comparable results with current state-of-the-art methods (whole tumor Dice segmentation: 0.89). The vascular heterogeneity assessment service was validated in a retrospective cohort of 50 patients, in a study focused on predicting patient overall survival based on the HTS habitats. Cox proportional hazard regression analysis and Kaplan-Meier survival study showed significant positive correlations (p-value < .05) between the HTS habitats and patient overall survival. ONCOhabitats system also generates radiological reports for each service, including volumetries and perfusion measurements of the different regions of the lesion.

Conclusion: ONCOhabitats system provides open-access services for glioblastoma heterogeneity assessment, implementing consolidated state-of-the-art techniques for medical image analysis. Additionally, we also give access to the scientific community to our computational resources, offering a computational capacity of about 300 cases per day.