PhD, University of Cincinnati, 2024, Medicine: Biomedical Informatics

The past decade has experienced an exponential development at the intersection of deep learning based artificial intelligence (AI) and medical imaging in healthcare. This synergistic association has primarily taken the shape of computer-aided decision (CAD) systems in the healthcare industry. CAD systems are used to harness the power of AI for the purpose of detection and diagnosis in biomedical images, which reduces human engagement in repetitive tasks, lowers healthcare costs, improves performance and decreases variability for decision making. The most common AI tasks are object detection (triage, CADt), segmentation (detection, CADe), and classification (diagnosis, CADx). Biomedical images, comprising of radiological and microscopy images, pose unique issues of data paucity, patient heterogeneity, image quality, and concerns associated with retrospective data collection – imbalanced datasets, limitations of object instances, and varying imaging protocols. In this thesis we present four bodies of work that demonstrate challenges and opportunities in developing task-specific CAD systems for radiological and microscopy images. As part of this overarching goal, the first challenge addresses radiological volume compression. With rising demand for teleradiology, storing and transmitting 3D images without losing data is challenging. To this concern, we proposed an almost lossless and generalizable compression technique designed exclusively for 3D images, called 3D VOI-OMLSVD. It performs at par with industry standard models and has a fast decoding/decompression time. The second challenge was identifying potential models for segmentation, classification, and synthetic image generation. We used CNN and GAN models to predict the pediatric liver stiffness from corresponding ultrasound images and elastography values. When faced with issues concerning model overfitting and model not learning, we attempted to create synthetic data with generative AI. This data wa (open full item for complete abstract)

Committee: Surya Prasath Ph.D. (Committee Chair); Andrew Trout M.D. (Committee Member); Jonathan Dillman M.D. M.S. (Committee Member); Jasbir Dhaliwal M.B.B.S. M.R.C.P.C.H. M.Sc. (Committee Member) Subjects: Artificial Intelligence

Master of Engineering, Case Western Reserve University, 2024, Biomedical Engineering

Radiomic analysis of individual regions or acquisitions has shown significant potential for predicting treatment response to neoadjuvant therapy in rectal cancers via routine MRI. We present a novel multi-plane, multi-region radiomics framework for exploiting intuitive clinical and biological aspects of rectal tumor response on MRI. Using a multi-institutional cohort of 151 baseline T2-weighted axial and coronal rectal MRIs, 2D texture features were extracted from multiple regions of interest (tumor, tumor-proximal fat) across both axial and coronal planes, with machine learning analysis to identify descriptors predictive of complete response to neoadjuvant therapy. Our multi-plane, multi-region radiomics model was found to significantly outperform single-plane or single-region feature sets with a discovery area under the ROC curve (AUC) of 0.765±0.054, and hold-out validation AUCs of 0.700 and 0.759. This suggests multi- region, multi-plane radiomics could enable detailed phenotyping of treatment response on MRI and thus personalization of therapeutic and surgical interventions in rectal cancers.

Committee: Satish Viswanath (Committee Chair); Amit Gupta (Committee Member); Juhwan Lee (Committee Member) Subjects: Artificial Intelligence; Bioinformatics; Biomedical Engineering; Biomedical Research; Medical Imaging

Doctor of Philosophy, Case Western Reserve University, 2024, Biomedical Engineering

In this dissertation, novel artificial intelligence methods address clinical gaps in gynecological oncology. Pathological and molecular heterogeneity in ovarian (OC), cervical (CC), and endometrial cancers (EC) leads to a lack of generalizable tumor biomarkers. This complexity necessitates innovative approaches for precision medicine, particularly in risk assessment. Furthermore, the disproportionate impact of \gls{EC} on African Americans underscores an urgent need to understand the molecular and cellular mechanisms behind this disparity. Radiomics and pathomics involve extracting features from radiology and pathology images using computational techniques. This research builds on the premise that such analyses can enhance cancer understanding and management. It aims to develop comprehensive predictive and prognostic models. Key objectives include using immune cell patterns for detailed tumor analysis, investigating genomics for molecular variance, and applying radiomics for cancer risk stratification. A significant technical contribution of this work is a novel computational pathology method that quantifies geospatial patterns of tumor-infiltrating lymphocytes (TILs) and cancer cells in OC, CC, and \gls{EC} from H\&E slides, as a non-invasive, cost-effective analysis and personalized treatment. Additionally, we captured the holistic characteristics of the tumor microenvironment by analyzing textural patterns in CT scans. The application of these methods spans various aspects of gynecological oncology. This includes stratifying patients and identifying molecular variances in African and European American populations affected by \gls{EC}. Genomic analysis revealed two gene sets differing significantly between populations, impacting tumor stroma and histomorphometry. Our study discovered distinct morphological differences in tumor epithelial nests and stroma, including the spatial arrangement and colocalization of TILs and tumor cells, as determined by stromal (open full item for complete abstract)

Committee: Anant Madabhushi (Advisor) Subjects: Biomedical Engineering

Master of Sciences (Engineering), Case Western Reserve University, 2023, Biomedical Engineering

Recent studies have highlighted the pathophysiological significance of epicardial adipose tissue (EAT) in the development of heart failure (HF). Using EAT image features (hereafter fat-omics) extracted from low-cost (no-cost at our institution) CT calcium score (CTCS) images, we predicted HF onset. We segmented EAT using a modified version of our deep learning algorithm, DeepFat, edited cases for accuracy, and collected 87 hand-crafted features (fat-omics) including volume, spatial, thickness, and HU values, as elevated HU is thought to be an indicator of inflammation. We included readily available clinical and demographic features (e.g. age, sex, and BMI). We used a dataset of HF-enriched patients (N=1,988, HF: 5.13%) and a Cox model with stepwise feature reduction, trained on 80% and evaluated with 20% held-out testing. High risk features (e.g., mean EAT thickness, EAT mean HU, and smoking) were identified using univariate analyses. Fat-omics, clinical and demographic features predicted HF with C-index/2-year AUC of 72.7/71.8, respectively. For comparison, different models using BMI, EAT volume, pericardial sac volume and a combined set of clinical and demographic features gave training/testing C-index values of 59.7/58.8, 60.0/59.8, 61.0/59.2 and 68.6/67.5, respectively. Additionally, we evaluated the calcification Agatston score, which is used to predict atherosclerosis-related major adverse cardiac events. It yielded training/testing of 62.7/62.9. Fat-omics, clinical and demographic features also gave excellent stratification of patients into low- and high-risk groups using Kaplan-Meier plots with a net reclassification improvement (NRI) of 0.11 (p-value=0.024) as compared to EAT Volume alone. Our results demonstrate that EAT plus simple clinical and demographic features might be used to predicting HF onset.

Committee: David Wilson (Committee Chair); Juhwan Lee (Committee Member); Shuo Li (Committee Member) Subjects: Biomedical Engineering; Health Care Management; Statistics

Doctor of Philosophy, Case Western Reserve University, 2023, EECS - Electrical Engineering

Initiating triggers of atrial fibrillation (AF) often arise from the pulmonary veins, which are targeted in ablation for AF. Cardiac morphologic changes are associated with persistent AF and are seen in AF risk-allele animal knock-down studies. Radiomic-based modeling has advanced clinical diagnostics in various clinical fields though has seen little application in atrial fibrillation, a common condition associated with risk of stroke. Radiomic and clinical features may be associated with likelihood of recurrence of AF post-ablation and may identify a radio-genomic phenotype for high-risk AF alleles. This approach builds on previously associated radiomic features and imaging modalities. Subjects who had pre-ablation contrast Computed Tomography (CT) scans prior to first-time catheter ablation for AF were allocated to training and validating datasets and AF recurrence was determined at 1 year based on health record review. Morphometric features pertaining to the pulmonary veins and left atria were measured manually and sixteen combinations of feature selection and classifier methods were applied to assess concordance among methods. Each method was compared based on the resulting C statistics, and models based on clinical features were similarly developed. Initial analysis with 5-fold cross validation identified radiomic and clinical features associated with post-ablation AF recurrence, and independent validation on a dedicated validation cohort confirmed these findings. We analyzed the influence of catheter ablation type and AF diagnosis type on pertinent features and identified right-sided pulmonary vein angles and left atrium volume normalized to height were associated with recurrence. Clinical features predictive of AF recurrence included older age, BMI, hypertension, and warfarin use. Models trained on radiomic and clinical features predicted AF recurrence with C-statistic of 0.76 (IQR 0.70-0.79). Radiomic and clinical features associated with increased l (open full item for complete abstract)

Committee: Christian Zorman (Committee Chair); Anant Madabhushi (Committee Member); Francis Merat (Committee Member); David Kazdan (Committee Member) Subjects: Electrical Engineering; Health Care; Medical Imaging; Medicine

Doctor of Philosophy, Case Western Reserve University, 2023, Biomedical Engineering

Current non-invasive cross-sectional imaging such as MRI and CT provide clinicians with a powerful tool for the diagnosis, monitoring, and treatment planning of patients. The advancements in this field are especially noticeable in chronic diseases like Crohn's, which benefits from early identification but lacks a long term cure and thus requires life time monitoring. However, the current application of imaging is predominantly qualitative, allowing for inter-reader variability and an incomplete view of the diseased regions. Additionally, many patients show significant variation in response to specific therapies with standard radiological and clinical assessment being unable to predict or prognosticate the response for each patient. This variation in response maybe due to underlying phenotypic differences in disease like extent of fibrosis and inflammation, but at present there are no methods to assess this at time of diagnosis with non-invasive imaging. However, the computer-extraction of advanced features from radiographic images (radiomics), has enabled superior disease characterization especially in concert with radiological reading and clinical markers. While, studies have shown the potential for radiomics in the context of treatment response prognostication in cancers there is a lack of similar work in the field of Crohn's. These initial studies have also shown that radiomic features vary as a function of the scanner and the settings. It is therefore important to identify a stable set of radiomic features which are correlated with Crohn's disease treatment outcomes and phenotype. In this dissertation, we provide a comprehensive evaluation of radiomic features in which, we identify pools of radiomic features which are consistent across image variations in both MRI and CT. We leverage these features to construct a prognostic radiomics model for risk stratifying patients with Crohn's based on need for early surgical interventions. Finally, we identify and validate (open full item for complete abstract)

Committee: David Wilson (Committee Chair); Satish Viswanath (Advisor); Anant Madabhushi (Committee Member); Shuo Li (Committee Member); Erick Remer (Committee Member) Subjects: Biomedical Engineering; Biomedical Research

Doctor of Philosophy, Case Western Reserve University, 0, Molecular Medicine

Novel therapeutic regimens which spur the endogenous immune system to kill cancer cells, such as stereotactic radiosurgery (SRS) and immune checkpoint inhibition (ICI), are heterogeneously effective. Understanding causal factors of response is vital to guide risk assessment and treatment decisions. In this thesis, I evaluate the ability of three methods to prognosticate survival for brain metastases patients following SRS and ICI treatment. These include the clinically utilized response assessment in neuro-oncology for brain metastases (RANO-BM) protocol, as well as investigational computational methods such as radiomic feature analysis and convolutional neural network (CNN) image analysis. I find that easing the 10mm RANO-BM diameter threshold for measurable disease allows new lesions to be discovered as proof of progression in ICI-treated metastases. Further, I find that the trajectory of RANO-BM diameter can be more instructive for risk prediction than the ratio-change and that neither volume nor number of metastases, nor RANO-BM diameter can significantly predict survival until a year after treatment. Reproducing common radiomic methodology flaws observed in the published literature, I demonstrate that inconsistent partitioning, or the improper division of radiomic feature data into Training, Validation, Test, and External test sets, can provide a 1.4x performance boost to reported accuracy (AUROC) for predictive models. Additionally, I highlight how spurious correlations with biological variables can overstate the importance of radiomic features. Leveraging the conclusions from my radiomic reproduction study, I assess the ability of radiomic features and convolutional neural networks (CNNs) to predict overall survival in the largest ICI-treated brain metastases cohort assembled to date, comprising 175 patients from three institutions in two countries. I find that neither radiomic features nor any architecture of the survival AI model MetsSurv is capable of p (open full item for complete abstract)

Committee: Jacob Scott (Advisor); Brian Rubin (Committee Chair); Elizabeth Gerstner (Committee Member); Anant Madabhushi (Committee Member); Jayashree Kalpathy-Cramer (Advisor); Nathan Pennell (Committee Member) Subjects: Artificial Intelligence; Computer Science; Immunology; Medical Imaging; Molecular Biology; Neurology; Oncology; Radiology

Doctor of Philosophy, Case Western Reserve University, 2022, Biomedical Engineering

Non-small cell lung cancer tumors follow an orderly progression from adenocarcinoma in situ (AIS) to minimally invasive carcinoma (MIA) and invasive adenocarcinoma (INV). Currently, there is no definite biomarker to access the level of invasion and detect invasive disease in these early lepidic lesions using radiographic scans, which would ideally help in surgery planning for these patients. Within the early-stage NSCLC cohort, while all the patients will receive the surgery, a significant portion of patients (up to 50\%) will develop recurrence. Although most of these patients are eligible to receive adjuvant chemotherapy (chemo), not all patients will receive the added benefits. In the more advanced NSCLC setting, immunotherapy (IO) has shown promising survival improvement, but only a fraction (20\%) of patients will respond to IO, and a fraction of patients (8\%) would, in fact, receive adverse effects of it, and cancer would spread rapidly (hyperprogression). Most of the current AI methods developed in this field are based on a single modality. However, information across different modalities and scales may hold complementary information, and integrating them together may enhance the performance of AI models. In addition, most of the developed AI models lack interpretability, an essential element for successfully transitioning these AI methods into clinical practices. In this dissertation, we introduced new interpretable AI biomarkers that use textural patterns on radiographic scans, known as Radiomics, and combine these biomarkers across multiple modalities and scales for NSCLC and COVID-19 patients. The Radiomic features were analyzed from inside the tumor region as well as from the area immediately surrounding the nodule. Furthermore, we integrated the clinical features into Radiomics Model by using novel techniques. We also created a human-machine integrated model using Radiologists' scores combined with Radiomic Analysis. Lastly, we used pathology data (open full item for complete abstract)

Committee: Anant Madabhushi (Advisor) Subjects: Biomedical Engineering; Biomedical Research

Master of Sciences, Case Western Reserve University, 2022, Biomedical Engineering

Evaluating tumor regression and nodal metastasis in rectal cancers via MRI after standard-of-care chemoradiation therapy (CRT) remains highly challenging in clinical practice. Recent studies have shown that physiologic environments surrounding tumor regions may provide complementary information that is predictive of response to CRT and patient survival. In order to capture physiologic changes due to CRT within the rectal wall (primary location of tumor) and perirectal fat (region surrounding tumor), we evaluated the performance of radiomics characterization of these regions within the rectal environment on post-chemoradiation T2-weighted (T2w) MRI in predicting tumor regression and nodal response after CRT. A total of 83 rectal cancer patients for whom MRIs as well as pathologic tumor staging were available post-CRT were included in this study, from two different institutions. Region-wise radiomic features were extracted from expert annotated perirectal fat and rectal wall regions on MRI and a 2-stage feature selection was employed to identify the most relevant features from each of these locations. Combining radiomic features from across the fat and wall yielded significantly improved classifier performance for identifying tumor stage regression (AUC=0.86) as well as nodal metastasis (AUC = 0.7) when compared to the performance of these regions individually; a trend that generalized across discovery and hold-out validation. Top-ranked radiomic features were also found to be associated with pathologic complete response, surrogates of long-term survival, as well as revealed significant differences between age- and sex-specific subgroups.

Committee: Satish Viswanath (Committee Chair) Subjects: Biomedical Engineering

Doctor of Philosophy, Case Western Reserve University, 0, Biomedical Engineering

Glioblastoma multiforme (GBM) is an aggressive, grade IV brain cancer. The current standard-of-care treatment for GBM patients is multimodal that includes surgical resection followed by radiotherapy and concomitant chemotherapy with temozolomide i.e. chemoradiation-therapy (CRT). However, in spite of such an aggressive treatment, GBM patients have a dismal median survival of 12-15 months and only <10% patients survive for over 5 years. Unfortunately, over 40% of GBM patients undergo disease progression within few weeks of CRT treatment. This poor prognosis can be attributed to genetic instability and intra- and inter-tumor heterogeneity of GBM that leads to treatment resistance, progression, and tumor recurrence. Although, isocitrate dehydrogenase-1 (IDH1) mutations, O6-methylguanine-DNA methyltransferase (MGMT) promotor methylation, and extent of resection have shown promise as prognostic biomarkers, to our knowledge, currently there are no validated image-based biomarkers that could apriori determine the risk of poor survival in a GBM patient. Each patient is unique with distinct morphological as well as phenotypical profiles. There is thus an unmet need to identify non-responder patients prior to CRT treatment and to predict progression-free survival, for personalizing treatment decisions in GBM patients. Radiographic imaging such as magnetic resonance imaging (MRI) is routinely used for clinical diagnosis and response assessment of GBM by manual visual inspection. Similarly, surgically resected tissue slides contain rich phenotypic information that could reveal the inherent intra-tumoral heterogeneity and thus has prognostic implications. Recent advances in computational techniques such as radiomics and pathomics have shown improved efficacy (over manual inspection) for prognosis and response assessment of GBM tumors from MRI scans and histopathology slides, respectively. However, there still remain a few open questions that need to be addressed in order to b (open full item for complete abstract)

Committee: Pallavi Tiwari (Advisor); Anant Madabhushi (Committee Chair); Manmeet Ahluwalia (Committee Member); Efstathios Karathanasis (Committee Member); Jennifer Yu (Committee Member) Subjects: Artificial Intelligence; Biomedical Engineering; Computer Engineering; Computer Science; Health Care; Medical Imaging

Master of Science in Biomedical Engineering, Cleveland State University, 2021, Washkewicz College of Engineering

The usage of radiomics for extracting high-dimensional features from radiographic imaging to quantify subtle changes in tissue structure and heterogeneity has shown great potential for disease diagnosis and prognosis. However, radiomic features are known to be impacted by acquisition-related changes (e.g., dose and reconstruction variations in CT scans) as well as technical variations between cohorts (i.e., batch effects due to varying dosage and tube currents). Using features which are not resilient to such imaging variations can result in poor performance of the downstream radiomics classifier models. In this study, we present a framework to systematically identify radiomic features that are resilient to both batch effects and acquisition differences, as well as evaluate the impact of such variations on radiomic model performance. We demonstrate the utility of our approach in the context of distinguishing active Crohn's disease (CD) from healthy controls using a uniquely accrued cohort of 164 CTE scans accrued from a single institution, which included (a) batch effects due to variations in effective dosage and tube current, as well as (b) scans simultaneously acquired at multiple doses and reconstructions (3 variations per patient). Our framework involves systematically evaluating the impact of acquisition variations (based on feature robustness to explicit dose/acquisition changes) and batch effects (based on feature stability to implicit dosage/current variations). Resilient radiomic features identified after accounting for both types of variations yielded the best random forest classifier performance across both discovery (AUC=0.819 ± 0.043) and validation (AUC=0.787) cohorts when using full-dose images; also found to be significantly more generalizable than features that were not optimized for such variations (AUC=0.419 in validation). This subset of radiomic features that were both robust and stable (resilient) also maintained their performance when evaluate (open full item for complete abstract)

Committee: Satish E. Viswanath (Committee Chair); Hongkai Yu (Committee Member); Moo-Yeal Lee (Advisor) Subjects: Biology; Biomedical Engineering; Biomedical Research; Medical Imaging; Radiology

Doctor of Philosophy, Case Western Reserve University, 2020, Biomedical Engineering

Personalized interventions in ongologic applications could enable significant benefits to cancer patients by tailoring treatment based on disease phenotype and associated treatment response. Not only will such strategies yield better outcomes, but also will incur lower morbidity rates and better quality-of-life by avoiding unnecessary operations. In the context of locally advanced rectal cancer, current standard-of-care recommends surgical removal of the entire rectum and surrounding tissue after neoadjuvant chemoradiation. However, a significant proportion of rectal cancer patients exhibit minimal to no remaining disease on the excised specimen, and could have been candidates for non-surgical, active surveillance. Magnetic resonance imaging (MRI) is routinely utilized for staging and re-staging the tumor before and after treatment, but is subjective and known to have poor correlation with pathologic staging. Recently, the computerized-extraction of more advanced features from radiographic images, or radiomics, which attempts to quantifying tissue attributes on imaging, has enabled improved disease characterization compared to visual inspection alone. While initial studies have shown promise for radiomic analyses in the context of predicting treatment response for rectal cancers, there is still a need to understand what they are capturing and ensure their utility between different scanners and different hospitals. Traditionally, statistical descriptors are used to describe the heterogeneity captured by radiomic operators, but may be sensitive to noise and not adequately characterize treatment-related changes in the rectal environment. It is therefore important to design radiomic features which capture more information that is pathologically or physiologically intuitive. Finally, in order to ensure their utility in a clinical setting, it is important to identify a stable set of radiomic features, or features which are reproducible across different sites and consistent in n (open full item for complete abstract)

Committee: Anant Madabhushi PhD (Committee Chair); Satish Viswanath PhD (Advisor); David Wilson PhD (Committee Member); Sharon Stein MD (Committee Member); Andrei Purysko MD (Committee Member) Subjects: Biomedical Engineering; Computer Science; Oncology

Doctor of Philosophy, Case Western Reserve University, 2020, Biomedical Engineering

At least 34% of prostate cancer (PCa) patients are diagnosed with low-risk disease, which makes them prospective candidates for active surveillance (AS) where these patients are routinely monitored via repeated Prostate-Specific Antigen measurements and biopsies for signs of disease progression. Prostate multiparametric-MRI (mpMRI) provides structural and functional measurements on multiple sequences for lesion characterization. It is widely used for diagnosing PCa which leads to the development of standardized guidelines for acquiring, reading and reporting findings through the Prostate Imaging-Reporting and Data System (PI-RADS). MpMRI has been proposed as a means of non-invasive monitoring of disease progression for patients on AS, yet, its inability to distinguish disease from tumor confounders and the inter-observer variability in terms of interpretation limits its complete adoption. Accurately classifying PCa tumors in addition to providing quantitative assessment of differences between AS patient groups who express different levels of PCa progression would not just obviate the need for biopsies and all the side effects that come with them, but also would pave the road for finding reliable solutions for the overdiagnosis and overtreatment of prostate cancer in the US. Radiomic analysis is the use of computer-extracted features for quantitatively characterizing subtle, sub-visual patterns associated with PCa. Features capturing intra- and peri-tumoral heterogeneity have been shown to distinguish low- from high-risk PCa. The sub-visual patterns captured by these features are not usually discernible to radiologists and hence potentially could add complementary information to the visual interpretation of the MRI. Common examples of radiomic features are first order statistics (mean, median, standard deviation, skewness, and kurtosis), grey-level co-occurrence (Haralick, Collage) and Laws' energy. In this work, radiomic analysis of mpMRI allowed for 1) discrimin (open full item for complete abstract)

Committee: Anant Madabhushi PhD (Committee Chair); David Wilson PhD (Committee Member); Lee Ponsky MD (Committee Member); Michael Kattan PhD (Committee Member) Subjects: Artificial Intelligence; Biomedical Engineering; Medical Imaging

Master of Sciences, Case Western Reserve University, 2020, Biomedical Engineering

Genomic characterization of Medulloblastoma (MB), an aggressive pediatric tumor, has recently identified 4 distinct molecular subgroups: Sonic Hedgehog (SHH) , Wingless (WNT) , Group 3, Group 4 each exhibiting different clinical behavior. The molecular sub-types have unique risk-profiles and outcomes, and patients could potentially benefit from sub-group specific treatments. However, the transition of these molecular MB sub-types into clinical practice has been limited due to challenges in availability of molecular profiling in most clinics. The hypotheses we sought to examine in this preliminary study was whether computer extracted deformation features of medulloblastomas from T1-weighted MRI could independently (1) distinguish between molecularly determined subgroups, and (2) distinguish between high risk and low risk MB patient populations based on their overall survival. Our feasibility results suggest that the subtle tissue deformation features in the brain around tumor region on routine MRI may potentially serve as surrogate markers to non-invasively characterize molecular sub-types, as well as to predict survival risk in pediatric MB patients.

Committee: Pallavi Tiwari PhD (Advisor); Anant Madabhushi PhD (Committee Member); David Wilson PhD (Committee Member); Benita Tamrazi MD (Committee Member) Subjects: Biomedical Engineering

Doctor of Philosophy, Case Western Reserve University, 2020, Biomedical Engineering

As the arsenal of therapeutic strategies in the fight against cancer grows, so too does the need for predictive biomarkers that can precisely guide their use in order to match patients with their optimal personalized treatment plan. Currently, clinicians often have little recourse but to initiate treatment and monitor a tumor for signs of response or progression, which exposes non-responsive patients to overtreatment, harmful side effects, and windows of ineffective therapy that increase a patient's risk of progression or metastasis. Thus, there is an urgent need for new sources of predictive biomarkers to help more effectively plan personalized treatment strategies. Radiological images acquired before treatment may contain previously untapped predictive information that can be quantified in the form of computational imaging biomarkers. The vast majority of existing computational imaging biomarkers provides analysis limited to the tumor region itself. However, the tumor environment contains critical biological information pertinent to tumor progression and treatment outcome, such as tumor-associated vascularization and immune response. This dissertation focuses on the development of new, biologically-inspired computational imaging biomarkers targeting the tumor environment for the prediction of response to a wide range of chemotherapeutic and targeted treatment strategies in oncology. First, we explore measurements of textural heterogeneity within the tumor and surrounding peritumoral environment, and demonstrate the ability to predict therapeutic response and tumor biology to neoadjuvant chemotherapy in primary and targeted therapy in primary and metastatic breast cancer. Second, we introduce morphologic techniques for the quantification of the twistedness and organization of the tumor-associated vasculature, and demonstrate their association with response and survival following four different therapeutic strategies in breast cancer MRI and non-small cell lung canc (open full item for complete abstract)

Committee: Madabhushi Anant (Advisor); Wilson David (Committee Chair); Abraham Jame (Committee Member); Gilmore Hannah (Committee Member); Plecha Donna (Committee Member); Varadan Vinay (Committee Member) Subjects: Biomedical Engineering; Biomedical Research; Computer Science; Medical Imaging; Medicine; Oncology; Radiology

Master of Sciences (Engineering), Case Western Reserve University, 2018, Biomedical Engineering

Radiomics has enabled the development of a number of prognostic and predictive imaging based tools, but there is limited work on benchmarking radiomic features across multiple sites and scanners; especially in the context of MRI. We benchmarked 5 radiomic feature families in terms of discriminability and reproducibility in a multi-site setting; specically attempting to characterize prostate tumors. 147 patient T2w MRI datasets from 4 different sites were pre-processed to correct for acquisition-related artifacts and differences. 406 3D radiomic features were extracted and evaluated across sites via quantitative measures of cross-site reproducibility and discriminability. We demonstrated that the majority of Haralick features were both highly reproducible and discriminable across sites. By contrast, Laws features were extremely variable in terms of both benchmarks. Our results indicate that only a subset of radiomic features and parameters may be generalizable enough for use in machine learning.

Committee: Satish Viswanath (Committee Chair); Anant Madabhushi (Committee Member); David Wilson (Committee Member); Andrei Purysko (Committee Member) Subjects: Biomedical Engineering

Doctor of Philosophy, Case Western Reserve University, 2017, Biomedical Engineering

Cancer is not a bounded, self-organized system. Most malignant tumors have heterogeneous growth, leading to disorderly proliferation well beyond the surgical margins. In fact, the impact of certain tumors is observed not just within the visible tumor, but also in the immediate peritumoral, as well as in seemingly normal-appearing adjacent field. Visual inspection is often not a reliable instrument in cancer diagnosis, providing only qualitative analysis of an image, thereby missing subtle disease signatures. These, and other imaging limitations can lead to unnecessary surgical interventions. Computerized image analysis has shown promise in comprehending disease heterogeneity through quantification and detection of sub-visual patterns. In this work, we present novel radiomic tools to identify subtle radiologic cues (radiomic descriptors) and address clinical challenges in cancer diagnosis, prognosis, and treatment-evaluation. The developed tools and techniques are modality- and domain-agnostic. They can be applied in a pan-cancer setting to mine information from radiographic images and discover associations with underlying molecular (radio-genomics) or histological (radio-pathomics) characteristics to provide a holistic characterization of disease. We have demonstrated their efficacy in addressing problems in prognosis and treatment management of brain tumors. The challenges we target specifically include (1) inability to estimate survival at a pre-treatment stage and (2) inability to avoid highly-invasive surgeries in patients with radiation-induced treatment changes that mimic tumor recurrence. Underlying heterogeneity is linked to poor prognosis and tumor recurrence. Cellular level differences associated with the distinct physiological pathways might also manifest at the radiographic (i.e. MRI) length scale. We present two radiomic descriptors, Co-occurrence of Local Anisotropic Gradient Orientations (CoLlAGe) and radiographic-Deformation and Textural Heterogenei (open full item for complete abstract)

Committee: Anant Madabhushi (Advisor); Pallavi Tiwari (Committee Chair); David Wilson (Committee Member); Lisa Rogers (Committee Member); Charles Lanzieri (Committee Member) Subjects: Biomedical Engineering; Biomedical Research

Master of Sciences (Engineering), Case Western Reserve University, 2017, Biomedical Engineering

Radiomics has shown promise for in vivo prediction of cancer risk, thus providing a potential avenue for reducing over-treatment and unnecessarily invasive biopsy-based diagnosis. Radiomics could be particularly beneficial for stratifying patients into different risk groups in the context of prostate cancer (PCa), for which limitations of current in vivo risk assessment result in over-diagnosis and over-treatment. Despite its promise, successful translation of radiomics into the clinic may require a more comprehensive understanding of the underlying morphologic tissue characteristics they reflect. Few studies, however, have attempted to establish the biological or histomorphometric basis for the performance of radiomics. Accomplishing this requires fusing the information obtained from the imaging modalities of radiology and histopathology, since the gold standard definition of PCa comes from histopathologic analysis of whole-mount specimens. The first step in performing this radiology-pathology fusion in PCa entails achieving spatial correspondence between preoperative in vivo magnetic resonance imaging (MRI) and ex vivo hematoxylin & eosin (H&E)-stained whole-mount radical prostatectomy specimens via deformable co-registration. Co-registration, however, requires whole-mount histology sections (WMHSs), which are not always feasible to obtain. In such cases, large specimens are cut into multiple smaller tissue fragments. This thesis presents work on two related modules of radiology-pathology fusion in PCa: First, a novel automated program called AutoStitcher, which reconstructs pseudo whole-mount histology sections (PWMHSs) by digitally stitching together multiple smaller tissue fragments, thus enabling co-registration with in vivo radiographic imagery. AutoStitcher reconstructed PWMHSs with less than 3% error relative to manually stitched PWMHSs. Second, comprehensive sets of radiomic features extracted from MRI and quantitative histomorphometric features from H (open full item for complete abstract)

Committee: Anant Madabhushi (Advisor); Satish Viswanath (Committee Member); David Wilson (Committee Member) Subjects: Biomedical Engineering; Computer Science; Engineering; Medical Imaging; Oncology; Radiology

Master of Sciences (Engineering), Case Western Reserve University, 2016, Biomedical Engineering

Medical imaging has become widespread for evaluating treatment response in vivo by measuring changes in tumor size or contrast uptake. However, the imaging data alone is limited as it does not capture all information needed for a complete characterization of tumoral response to treatment. Recently, the use of image analysis has enabled strategies for more intelligently extracting information from imaging data. However, these strategies have been scarcely used for treatment response evaluation purposes. In this work, we investigate the development of two classes of image analysis called (1) radiomics for early treatment response characterization and (2) radiologypathology fusion. Radiomics will enable the detection of more subtle changes in the tumor environment in vivo early into the treatment regime, and the fusion of radiology and pathology data will allow annotations of treatment effects seen on the pathology specimen to be spatially mapped onto the corresponding imaging.

Committee: Anant Madabhushi Dr. (Advisor); Pallavi Tiwari Dr. (Committee Member); Xin Yu Dr. (Committee Member) Subjects: Biomedical Engineering; Biomedical Research; Medical Imaging