Quantitative imaging and molecular data modeling for brain tumor recurrence and progression analysis

Old Dominion University, Norfolk VA

Investigators

Linked publications & trials

Abstract

Project Summary/Abstract The overall prognosis for Glioblastoma Multiforme (GBM) brain tumors is extremely poor with a median survival of about one year from diagnosis and five-year survival rates of only 6.9%. Multiple biological characteristics contribute to the lethality of GBM including rapid, uncontrolled proliferation throughout the restricted cranial space supported by their pro-angiogenic nature and ability to rapidly develop therapeutic resistance resulting in tumor recurrence (TR). Identification of early TR is complicated by chemoradiation induced toxicity which can present initially as pseudo-progression and later as radiation necrosis (RN) condition that occurs in 25% of all GBM. Currently, invasive stereotactic brain biopsy at the site of suspicion remains the only resort for TR confirmation. This proposed renewal study will develop novel computational models and AI methods for analyzing brain TR, radiation necrosis (RN) and postoperative rapid early progression (REP); and to rigorously validate the models using both in-house and public domain patient data in collaboration with the ReSPOND consortium. This goal will be accomplished via the following aims: (1) Multi-center multimodal patient data collection, data harmonization, and robust TR, edema and RN volume segmentation and tracking; (2) Co-analysis of imaging and molecular features for TR and progression analysis; (3) Explainable AI modeling with uncertainty analysis to understand TR and progression; and (4) Evaluation and cross-validation of the proposed methods. The proposed co-analysis of radiomics, proteomics, and histopathology data is expected to enable robust TR, progression and REP analysis that may be critical to stratification of aggressive versus non-aggressive brain tumors for patients, with the potential for targeted therapy transcending the current limitations of the one-size- fits-all radiation treatment planning paradigm. In our parent R01 project, we made considerable progress in: (i) addressing critical barriers related to brain tumor volume segmentation, tumor growth tracking, glioma diagnosis and grading; (ii) GBM patient survival prediction using multi-modal radiology, molecular and histopathology patient data; (iii) molecular prediction of diffuse low-grade glioma using advanced radiomics features; and (iv) prediction of low-grade glioma progression using radiomics features. Recently, we showed the feasibility of multiresolution radiomics texture features to discriminate TR from RN. Very recently, our pilot study showed the feasibility of quantitative co-analysis of radiomics, proteomic and clinical patient data for REP and patient survivability prediction. Furthermore, we have been consistently ranked among the top teams in Global Challenges on Brain Tumor Segmentation (BRaTS), Brain Ischemic Segmentation, and Patient Survivability Prediction since 2013. These advances will be foundational in achieving the aims of this proposed study.