Deep learning in paediatric neuroimaging
Research type
Research Study
Full title
Deep learning for paediatric neuroimaging
IRAS ID
272174
Contact name
Kshitij Mankad
Contact email
Sponsor organisation
Stanford University
Duration of Study in the UK
5 years, 0 months, 0 days
Research summary
Clinical outcomes in pediatric patients with brain tumors are highly variable even following state-of-the-art treatment. Current approaches to personalized pediatric brain tumor prognostics focus on advancing tumor molecular characterization, which can be costly, invasive, and require complex technologies not readily available to most hospitals. Therefore, there is a significant need for an easily accessible, non-invasive marker of pediatric tumor sub typing and prognosis. In this proposed study, we hypothesize that MRI-based computational features can be used to accurately and reproducibly predict pediatric brain tumor histology, molecular subtypes, as well as prognosis, as measured by progression free survival (PFS) and overall survival(OS). Our goal is to assemble a database of pediatric brain tumor MRI and clinical data from at least five institutions across the world. Using this database, we aim to: (1) identify quantitative image-based signatures for pediatric brain tumors compared to normal neural development and myelination changes of childhood; (2) develop an image-based model predictive of histological and molecular subtypes, as well as PFS and OS in various pediatric brain tumors and (3) evaluate model generalizability through validation in a multi-institutional cohort.
REC name
N/A
REC reference
N/A