AI-Enabled Evaluation of Treatment Response in Recurrent Brain Tumors Using MRI

NIH

Motivation & Problem Description

  • More than 90% of patients with glioblastoma—one of the most aggressive types of primary brain cancer—experience recurrence, often within months of initial treatment.
  • Distinguishing recurrent tumor from treatment-induced changes in the patient’s brain is essential for evaluating treatment response and planning subsequent therapy.
  • Biopsy can provide this distinction, but it is invasive and samples only a small portion of the lesion, while the two conditions can look very similar on MRI.

Objective

  • Use MRI and advanced AI to map recurrent tumor and treatment-induced changes throughout each patient’s brain.
  • Support individualized, accurate treatment evaluation and clinical decision-making.

Solution

  • Developed BioNet, a novel AI model integrating MRI and cancer biology, producing patient-specific, voxel-resolution prediction maps of recurrent tumor and treatment-induced reactive or inflammatory tissue.

Impact

  • Demonstrated robust performance across patient cohorts with different imaging conditions.
  • Could improve assessment of whether treatment is working.
  • Could enable noninvasive monitoring and more timely treatment adjustments. 
AI-Enabled Evaluation of Treatment Response in Recurrent Brain Tumors Using MRI
BioNet (middle), a biologically informed AI model, takes a patient’s multiparametric MRI as input (left) and generates voxel-level maps showing the spatial distributions of recurrent tumor and treatment-induced changes (right). These maps can help clinicians evaluate treatment effectiveness in recurrent glioblastoma and plan subsequent therapy. 
References:

Wang, H., Argenziano, M. G., Yoon, H., Boyett, D., Save, A., Petridis, P., Savage, W., Jackson, P., Hawkins-Daarud, A., Tran, N., Hu, L., Singleton, K.W., Paulson, L., AI Dalamah, O., Bruce, J.N., Grinband, J., Swanson, K.R., Canoll, P. & Li, J. (2024). Biologically informed deep neural networks provide quantitative assessment of intratumoral heterogeneity in post treatment glioblastoma. npj Digital Medicine, 7(1), 292.

Project PI

Jing Li

Jing Li

Associate Chair for Faculty Development and Research
Virginia C. and Joseph C. Mello Chair
Professor

ISyE ProfilePersonal Website

Project Details

Sponsors:

NIH

Research Specializations:

Analytics and Machine Learning Data Science and Statistics Health and Humanitarian Systems System Informatics and Control

Application Domains:

Healthcare

Collaborators:

Mayo Clinic, Columbia University Irving Medical Center