A Multimodality Multimodal Deep Learning Framework on MRI Imaging and Genomics to Assess Brain Cancer Survival

Identifier:  TFM:2114

Handle: https://hdl.handle.net/20.500.11797/TFM2114

Authors:  Buzdugan, Sebastian Eugen

Abstract:
Glioblastoma (GBM) is a cerebral neoplasm characterised by intricate genetic factors that impact the prognosis of patients. The primary objective of this work is to predict survival by combining important molecular changes, such as IDH1 mutations and MGMT promoter methylation, with MRI-derived radiomics and clinical data. Improved survival prediction in GBM patients was achieved by employing machine learning models such as RandomForest, XGBoost, LightGBM, and a bespoke Dense Neural Network. The Dense NN model achieved higher accuracy when trained on UPENN-GBM (602 patients) and UCSF-PDGM (414 patients). This strategy emphasizes the capacity of artificial intelligence and radiogenomics to enhance precision medicine for glioblastoma multiforme (GBM).