Advancing Brain Tumor Analysis: Curating a High-Quality MRI Dataset for Deep Learning-Based Molecular Marker Profiling

Divya D. Reddy, Niloufar Saadat, James M. Holcomb, Benjamin C. Wagner, Nghi C. Truong, Jason Bowerman, Kimmo J. Hatanpaa, Toral R Patel, Marco C. Pinho, Ananth J Madhuranthakam, Chandan Ganesh Bangalore Yogananda, Joseph A. Maldjian; Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) Workshops, 2024, pp. 2373-2379

This study sought to address the critical need for high-quality datasets to advance deep learning (DL) models for non-invasive profiling of brain tumor molecular markers. We curated a comprehensive brain tumor dataset from patients diagnosed at University of Texas at Southwestern Medical Center (UTSW) between 2012 and 2020. The study curated 1740 MRI sessions from 709 subjects emphasizing pre-operative cases. Molecular analyses were conducted on tumor tissues as part of standard clinical care to identify key genetic alterations (MGMT IDH 1p/19q co-deletion). Detailed demographic histological and molecular marker information were extracted from the clinical electronic medical record (EMR). An internal reporting system was created in XNAT to comprehensively catalogue and curate the imaging and associated clinical data. MR Images were pre-processed using the FeTS platform and quality checked (QC) by expert neuroradiologists. Brain tumors were segmented into key components using FeTS and were manually corrected by experienced research staff and verified by expert neuroradiologists. Here we describe the critical challenges approaches and solutions for creating a high-quality curated medical imaging dataset for use in deep learning studies.