Date of Award

Summer 8-2020

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Electrical/Computer Engineering

Program/Concentration

Electrical Engineering

Committee Director

Khan M. Iftekharuddin

Committee Member

Chunsheng Xin

Committee Member

Jiang Li

Committee Member

Norou Diawara

Abstract

The work in this dissertation proposes model-based approaches for molecular mutations classification of gliomas, grading based on radiomics features and genomics, and prediction of diffuse gliomas clinical outcome in overall patient survival. Diffuse gliomas are types of Central Nervous System (CNS) brain tumors that account for 25.5% of primary brain and CNS tumors and originate from the supportive glial cells. In the 2016 World Health Organization’s (WHO) criteria for CNS brain tumor, a major reclassification of the diffuse gliomas is presented based on gliomas molecular mutations and the growth behavior. Currently, the status of molecular mutations is determined by obtaining viable regions of tumor tissue samples. However, an increasing need to non-invasively analyze the clinical outcome of tumors requires careful modeling and co-analysis of radiomics (i.e., imaging features) and genomics (molecular and proteomics features). The variances in diffuse Lower-grade gliomas (LGG), which are demonstrated by their heterogeneity, can be exemplified by radiographic imaging features (i.e., radiomics). Therefore, radiomics may be suggested as a crucial non-invasive marker in the tumor diagnosis and prognosis. Consequently, we examine radiomics extracted from the multi-resolution fractal representations of the tumor in classifying the molecular mutations of diffuse LGG non-invasively. The proposed radiomics in the decision-tree-based ensemble machine learning molecular prediction model confirm the efficacy of these fractal features in glioma prediction. Furthermore, this dissertation proposes a novel non-invasive statistical model to classify and predict LGG molecular mutations based on radiomics and count-based genomics data. The performance results of the proposed statistical model indicate that fusing radiomics to count-based genomics improves the performance of mutations prediction. Furthermore, the radiomics-based glioblastoma survival prediction framework is proposed in this work. The survival prediction framework includes two survival prediction pipelines that combine different feature selection and regression approaches. The framework is evaluated using two recent widely used benchmark datasets from Brain Tumor Segmentation (BraTS) challenges in 2017 and 2018. The first survival prediction pipeline offered the best overall performance in the 2017 Challenge, and the second survival prediction pipeline offered the best performance using the validation dataset. In summary, in this work, we develop non-invasive computational and statistical models based on radiomics and genomics to investigate overall survival, tumor progression, and the molecular classification in diffuse gliomas. The methods discussed in our study are important steps towards a non-invasive approach to diffuse brain tumor classification, grading, and patient survival prediction that may be recommended prior to invasive tissue sampling in a clinical setting.

DOI

10.25777/xge2-9f89

ISBN

9798678108234

Share

COinS