Brain Tumor Intraoperative Ultrasound Database

Overview

Predicting the survival of patients diagnosed with glioblastoma (GBM) is essential to guide surgical strategy and subsequent adjuvant therapies. Intraoperative ultrasound (ioUS) is a low-cost, versatile technique available in most neurosurgical departments. The images from ioUS contain biological information possibly correlated to the tumor's behavior, aggressiveness, and oncological outcomes. Today's advanced image processing techniques require a large amount of data. Therefore, the investigators propose creating an international database aimed to share intraoperative ultrasound images of brain tumors. The acquired data must be processed to extract radiomic or texture characteristics from ioUS images. The rationale is that ultrasound images contain much more information than the human eye can process. Our main objective is to find a relationship between these imaging characteristics and overall survival (OS) in GBM. The predictive models elaborated from this imaging technique will complement those already based on other sources such as magnetic resonance imaging (MRI), genetic and molecular analysis, etc. Predicting survival using an intraoperative imaging technique affordable for most hospitals would greatly benefit the patients' management.

The investigators plan to carry out a multicentre retrospective study of patients operated with GBM diagnosis between January 2018 and January 2020, in order to set the base for future prospective collection of patients. All cases with an ioUS study will be included. All patients must count with B-mode modality. After an pseudonymization process, the images will be uploaded to a private cloud server. Demographic, clinical, conventional radiological, and molecular variables (IDH, MGMT) will also be collected. OS will be defined as the time elapsed between the histopathological diagnosis and the patient's death. The acquired data must be processed to obtain a series of radiomic markers to perform the study. A pre-processing stage will be necessary (noise cleaning, despeckling, intensity normalization, filtering) to calculate radiomics measurements (histogram, volumetric, shape, texture, etc.). In the previous stage, a very high number of radiological features per subject will be calculated. Because the number of features is much higher than the data set, to avoid the curse of dimensionality, it will be necessary to reduce their number using feature selection and extraction techniques (standard in pattern recognition and radiomics) that allow choosing those characteristics (or transformations of them) that have greater discriminating power. A predictive model of survival will then be elaborated based on the features selected. Hypotheses Intraoperative ultrasound images in B-mode harbour tumor texture features correlated with overall survival in glioblastomas. Objectives: * To determine the relationship between the radiomic features of intraoperative ultrasound B-mode and overall survival in glioblastomas. * Develop a predictive survival model using the texture features with the highest discriminatory power. * Validate the model against an external dataset and compare it with currently available predictive models. * Build a data set that allows exploring various image harmonization techniques that allow the reproducibility of our predictions. * Establish an international cooperation network (BraTioUS-DB) whose objective will be to interchange ultrasound images and clinical data of patients operated on for a brain tumor prospectively from its creation and start-up.