High-grade gliomas are highly aggressive central nervous system (CNS) malignancies with high fatality rates if left untreated. There is currently a lack of reliable diagnostic tools to characterize the diffuse cell populations commonly found in these tumors. Here, we report that electrochemical impedance spectroscopy (EIS) can be used in an in vitro system to analyze changes in the impedance contributed by the extracellular matrix (ECM) of two glioblastoma cell lines: GBM 22 and GBM 115. EIS was more effective at resolving differences in impedance from GBM 115 cells than GBM 22 cells, which depended on both cell confluency and frequency. However, differences in impedance were more apparent from the supernatant when the cells were removed in both cell lines. Analysis of the PC12 and either of the GBM cell line co-cultures yielded highly statistically significant differences between all comparisons of cell confluencies and frequency steps. These results illustrate that EIS can be an effective instrument for characterizing the ECM surrounding glioblastoma cells, providing insight into the cellular behavior of these oncogenic cells.

Glioblastoma multiforme (GBM) is an aggressive brain cancer without effectivetreatment options, leaving patient survival rates extremely low. HDAC1 knockdown was found to initiate an invasive phenotype in vivo, particularly within the BT145 human glioma stem cell (hGSC) line. Analysis through RNA sequencing (RNA-seq) gene expression and regulatory networks found both CEBPβ, a known transcription factor (TF) involved in cellular invasion, and the STAT3 pathway, a notorious genetic component of GBM, were differentially expressed in BT145 hGSCs after HDAC1 knockdown. Furthermore, overlap of genes regulated by CEBPβ and STAT3 indicate the CEBPβ/STAT3 pathway may be involved in the observed BT145- specific invasive phenotype. The SYstems Genetics Network AnaLysis (SYGNAL) pipeline was applied to construct sex-specific gene regulatory networks from The Cancer Genome Atlas (TCGA) GBM patient expression data. Unique bicluster eigengenes were discovered separately for all, female, and male patients. Through the application of these bicluster eigengenes to a GBM cohort with multiparametric magnetic resonance imaging (mpMRI) localized biopsies, sex-specific associations between bicluster expression, mpMRI readout, and hallmarks of cancer were determined. Distinctive cancer functions were revealed transcriptionally through bicluster expression, and connected to a unique mpMRI feature. Specifically, SPGRC mpMRI indicated a strong signal for both immune hallmarks (evading immune detection and tumor-promoting inflammation). At the same time, MD mpMRI displayed a tendency toward sustained angiogenesis, possibly signaling the formation of new blood vessels. Uncovering each mpMRI feature’s underlying biological processes enables improved GBM diagnosis and treatment utilizing an individualized, non-invasive approach.

Semi-supervised learning (SSL) is sub-field of statistical machine learning that is useful for problems that involve having only a few labeled instances with predictor (X) and target (Y) information, and abundance of unlabeled instances that only have predictor (X) information. SSL harnesses the target information available in the limited labeled data, as well as the information in the abundant unlabeled data to build strong predictive models. However, not all the included information is useful. For example, some features may correspond to noise and including them will hurt the predictive model performance. Additionally, some instances may not be as relevant to model building and their inclusion will increase training time and potentially hurt the model performance. The objective of this research is to develop novel SSL models to balance data inclusivity and usability. My dissertation research focuses on applications of SSL in healthcare, driven by problems in brain cancer radiomics, migraine imaging, and Parkinson’s Disease telemonitoring.

The first topic introduces an integration of machine learning (ML) and a mechanistic model (PI) to develop an SSL model applied to predicting cell density of glioblastoma brain cancer using multi-parametric medical images. The proposed ML-PI hybrid model integrates imaging information from unbiopsied regions of the brain as well as underlying biological knowledge from the mechanistic model to predict spatial tumor density in the brain.

The second topic develops a multi-modality imaging-based diagnostic decision support system (MMI-DDS). MMI-DDS consists of modality-wise principal components analysis to incorporate imaging features at different aggregation levels (e.g., voxel-wise, connectivity-based, etc.), a constrained particle swarm optimization (cPSO) feature selection algorithm, and a clinical utility engine that utilizes inverse operators on chosen principal components for white-box classification models.

The final topic develops a new SSL regression model with integrated feature and instance selection called s2SSL (with “s2” referring to selection in two different ways: feature and instance). s2SSL integrates cPSO feature selection and graph-based instance selection to simultaneously choose the optimal features and instances and build accurate models for continuous prediction. s2SSL was applied to smartphone-based telemonitoring of Parkinson’s Disease patients.

Glioma is a devastating, invasive form of brain cancer with a 36-month median overall survival. The highest grade tumors, glioblastomas, have an even shorter prognosis of about 15 months. A glioma often requires an intense combination of treatments including surgery, chemotherapy, and radiotherapy, often resulting in very damaging side effects. Due to their sensitive location in the brain, which is often difficult to access because of the skull, gliomas are most often visualized using magnetic resonance imaging (MRI), a non-invasive imaging method. Because high grade gliomas (HGGs) are highly aggressive and recurrence is common, patients diagnosed with these tumors stand to significantly benefit from novel, advanced MRI techniques that can lead to better patient-specific tumor characterization and improved response assessment. Magnetic resonance elastography (MRE) is a MRI-based method that measures the mechanical properties of tissue, and has the potential to significantly enhance the ability to distinguish malignant vs. healthy brain tissue by determining spatial differences in physical stiffness. We investigated whether the addition of MRE to standard clinical glioma MRI protocols would provide a more accurate understanding of the extent of tumor invasion. Using routinely available T2-weighted and contrast enhancing T1-weighted clinical MRI images, the Swanson lab has developed the Proliferation-Invasion (PI) model of brain tumor growth. Using this model, we quantify the relative diffusion (D) and proliferation (�) of tumor cells as D/�. Clinical MRIs were segmented in order to parameterize the model and determine these tumor growth metrics for each patient in our retrospective study. Next, we compare these tumor growth metrics with MRE features of physical stiffness of malignant tissue to determine whether there are correlations with the PI model's kinetic parameters. We hypothesized that MRE stiffness measurements would be associated with the PI model of glioma growth and may provide additional patient-specific tumor characterization information useful for optimally choosing treatment and understanding treatment response. MRE has the potential to be a useful addition to the clinical management of glioma and be integral to further understanding tumor growth and invasiveness.