Smith, Kris A.

Abstract BACKGROUND An important challenge in radiomics research is reproducibility. Images are collected on different image scanners and protocols, which introduces significant variability even for the same type of image across institutions. In the present proof-of-concept study, we address the reproducibility issue by using domain adaptation – an algorithm that transforms the radiomic features of each new patient to align with the distribution of features formed by the patient samples in a training set. METHOD Our dataset included 18 patients in training with a total of 82 biopsy sample. The pathological tumor cell density was available for each sample. Radiomic (statistical + texture) features were extracted from the region of six image contrasts locally matched with each biopsy sample. A Gaussian Process (GP) classifier was built to predict tumor cell density using radiomic features. Another 6 patients were used to test the training model. These patients had a total of 31 biopsy samples. The images of each test patient were purposely normalized using a different approach, i.e., using the CSF instead of the whole brain as the reference. This was to mimic the practical scenario of image source discrepancy between different patients. Domain adaptation was applied to each test patient. RESULTS Among the 18 training patients, the leave-one-patient-out cross validation accuracy is 0.81 AUC, 0.78 sensitivity, and 0.83 specificity. When the trained model was applied to the 6 test patients (purposely normalized using a different approach than that of the training data), the accuracy dramatically reduced to 0.39 AUC, 0.08 sensitivity, and 0.61 specificity. After using domain adaption, the accuracy improved to 0.68 AUC, 0.62 sensitivity, and 0.72 specificity. CONCLUSION We provide candidate enabling tools to address reproducibility in radiomics models by using domain adaption algorithms to account for discrepancy of the images between different patients.

Abstract INTRODUCTION The quantification of intratumoral heterogeneity – through radiomics-based approaches - can help resolve the regionally distinct genetic drug targets that may co-exist within a single Glioblastoma (GBM) tumor. While this offers potential diagnostic value under the paradigm of individualized oncology, clinical decision-making must also consider the degree of uncertainty associated with each model. In this study, we evaluate the performance of a novel machine-learning (ML) algorithm, called Gaussian Process (GP) modeling, that can quantify the impact of multiple sources of uncertainty in ML model development and prediction accuracy, including variabilities in the copy number measurement, radiomics features, training sample characteristics, and training sample size. METHOD We collected 95 image-localized biopsies from 25 primary GBM patients. We coregistered stereotactic locations with preoperative multi-parametric MRI features (conventional MRI, DSC perfusion, Diffusion Tensor Imaging) to generate spatially matched pairs of MRI and copy number variants (CNV) for for each biopsy. We developed a Gaussian Process (GP) model to predict CNV for Epidermal Growth Factor Receptor (EGFR) based on MRI radiomic features in each patient. We used leave-one-patient-out cross validation to quantify prediction accuracy and model uncertainty. Spatial prediction and uncertainty (p-value) maps were overlaid to help visualize regional genetic variation of EGFR and uncertainty of the radiomic predictions. RESULT: The initial GP radiomics model for EGFR amplification (CNV > 3.5) produced a sensitivity of 0.8 and specificity of 0.8. Samples/regions associated with high uncertainty (p-value >0.05) correlated with either 1) extrapolation of radiomic features from the training set-defined feature space or 2) insufficient training samples in the feature space. CONCLUSION We present a ML-based model that quantifies spatial genetic heterogeneity in GBM, while also estimating model uncertainties that result from multi-source data variabilities. This approach lays the groundwork for prospective clinical integration of modeling-based diagnostic approaches in the paradigm of individualized medicine.

Abstract Purpose: Investigate the utility of patient-specific spatial predictions of tumor cell density from a bio-mathematical model. Introduction: Glioblastomas (GBMs) are the most malignant of all primary brain tumors. While it is known there is always a non-detectable portion of the tumor, current techniques of monitoring GBM progression, imaging and initial histological assessment, are not able to reliably estimate the tumor invasion past the enhancing region on T2-Weighted (T2W) imaging. Over the last two decades, a large effort has been made to create a simple patient-specific mathematical model of gliomas. The resulting model, referred to as the Proliferation-Invasion (PI) model, is based on two key parameters, the net growth rate, ρ, and the dispersal coefficient, D. In this model, the ratio of D/ρ is related to degree of invasion and the product D*ρ, is related to the speed of growth. The intuitive understanding provided by this model has been able to provide patient-specific understanding of disease kinetics enabling prediction of outcomes following surgical resection, radiation and the development of a prognostic response metric. Previous literature utilizing this model has been based on the assumption that what is seen on the pretreatment T1-Weighted contrast-enhanced (T1Gd) and T2W, images correspond to an 80% and 16% tumor cell density threshold respectively. This assumption allows for an estimate of D/ρ from a single time point of imaging. While these values were based on extensive experience, for ethical and technical reasons, they have never been rigorously investigated histologically. Recent technological advances have made it possible for surgeons to use an MRI to guide the acquisition of tissue making it possible to know with a good degree of accuracy where on the MR image the histological specimen comes from. Methods: Model Calibration : To estimate D/ρ for each patient, we assume abnormalities on the T1Gd and T2W images correspond to an 80% and 16% tumor cell density threshold respectively. We then utilize a Bayesian calibration approach based on adaptive grid refinement while holding the velocity constant to find the most likely value of D/ρ to match the observed radial measurements. Three-Dimensional Density Maps : Given a gray/white segmentation and an estimate for D/ρ, we can build a tumor cell density prediction in the patient's anatomy using the Eikonal equations and the modified Fast Marching Method (FMM) algorithm presented by Konukoglu et al. Patient Cohort : Eighteen patients were recruited with clinically suspected GBM undergoing preoperative stereotactic MRI for surgical resection with IRB approval Barrow Neurological Institute and Mayo Clinic in Arizona. Surgical Biopsy : Pre-operative conventional MRI, including T1Gd and T2W, was utilized to guide stereotactic biopsies. An average of 5–6 tissue specimens were acquired from each tumor by using stereotactic surgical localization, following the smallest possible diameter craniotomies to minimize brain shift. Histological Analysis : 4 μm tissue sections were stained with hematoxylin and eosin (H&E) for neuropathology review. H&E slides were reviewed by a neuropathologist. The percent tumor nuclei in the field was estimated. Results and Conclusion: Our analysis showed that the intuitive ordering of diffuse to nodular tumor profiles from the PI model is consistent with the histologic data. Further, utilizing the histologic data, limitations to a universal threshold cutoff of the T2/FLAIR region are specified and a better threshold cutoff value is suggested. We hope that this paper will lead to more studies of this kind resulting in more accurate ways to predict the spatial distribution of GBM tumor cells from medical imaging. Such advances could revolutionize surgical procedures and radiation therapy and may enable additional insights into tumor kinetic differences meaningful for treatment. Citation Format: Andrea Hawkins-Daarud, Lauren DeGirolamo, Joshua Jacobs, Kamala Clark-Swanson, Jennifer M. Eschbacher, Kris A. Smith, Peter Nakaji, Leslie C. Baxter, John P. Karis, Teresa Wu, J. Ross Mitchell, Jing Li, Leland Hu, Kristin R. Swanson. Histologic evidence for a bio-mathematical model of glioblastoma invasion. [abstract]. In: Proceedings of the AACR Special Conference on Engineering and Physical Sciences in Oncology; 2016 Jun 25-28; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2017;77(2 Suppl):Abstract nr A08.

AbstractBackgroundGlioblastoma (GBM) exhibits profound intratumoral genetic heterogeneity. Each tumor comprises multiple genetically distinct clonal populations with different therapeutic sensitivities. This has implications for targeted therapy and genetically informed paradigms. Contrast-enhanced (CE)-MRI and conventional sampling techniques have failed to resolve this heterogeneity, particularly for nonenhancing tumor populations. This study explores the feasibility of using multiparametric MRI and texture analysis to characterize regional genetic heterogeneity throughout MRI-enhancing and nonenhancing tumor segments.MethodsWe collected multiple image-guided biopsies from primary GBM patients throughout regions of enhancement (ENH) and nonenhancing parenchyma (so called brain-around-tumor, [BAT]). For each biopsy, we analyzed DNA copy number variants for core GBM driver genes reported by The Cancer Genome Atlas. We co-registered biopsy locations with MRI and texture maps to correlate regional genetic status with spatially matched imaging measurements. We also built multivariate predictive decision-tree models for each GBM driver gene and validated accuracies using leave-one-out-cross-validation (LOOCV).ResultsWe collected 48 biopsies (13 tumors) and identified significant imaging correlations (univariate analysis) for 6 driver genes: EGFR, PDGFRA, PTEN, CDKN2A, RB1, and TP53. Predictive model accuracies (on LOOCV) varied by driver gene of interest. Highest accuracies were observed for PDGFRA (77.1%), EGFR (75%), CDKN2A (87.5%), and RB1 (87.5%), while lowest accuracy was observed in TP53 (37.5%). Models for 4 driver genes (EGFR, RB1, CDKN2A, and PTEN) showed higher accuracy in BAT samples (n = 16) compared with those from ENH segments (n = 32).ConclusionMRI and texture analysis can help characterize regional genetic heterogeneity, which offers potential diagnostic value under the paradigm of individualized oncology.

Abstract Purpose: This study uses spatially accurate image-guided tissue analysis to evaluate the correlation between Perfusion MRI (pMRI) and tissue microvascular density parameters as markers for tumor recurrence and treatment response in high-grade glioma (HGG). Methods: Following Institutional Review Board approval, we recruited previously treated (including chemo-radiation therapy) HGG patients undergoing surgical re-resection of new enhancing lesions on surveillance MRI. Preoperatively, we acquired pMRI and contrast-enhanced stereotactic T1-Weighted (T1W) MRI data sets. Intraoperatively, we recorded stereotactic locations of multiple biopsies based on T1W data sets that guided surgery. Following surgery, we coregistered pMRI and T1W data sets to calculate localized relative cerebral blood volume (rCBV) for each stereotactic specimen location. For each specimen, we performed 1) standard H&E staining to diagnose tumor recurrence or post-treatment radiation effect (PTRE); and 2) immunohistochemical analysis with CD34 to highlight tissue vessels. We analyzed CD-34 stained slides with Axiovision Automeasure 3.4 (Zeiss, Germany) to calculate both total microvessel number (MVN) and total microvessel area (MVA), each normalized to total slide specimen area (μm2). We calculated Pearson correlations to establish relationships between a) rCBV and MVN; and b) rCBV and MVA. We also performed t-test to compare MVN, MVA, and rCBV values between tumor and PTRE samples. A neuroradiologist performed all coregistration and pMRI calculations, and a neuropathologist analyzed all tissue specimens, without knowledge of corresponding data. A biostatistician performed all statistical comparisons. Results: In this preliminary study, we included 16 tissue specimens (from 8 subjects), each diagnosed as either tumor (n=7) or PTRE (n=9). We successfully calculated localized rCBV and determined both MVA and MVN for each specimen. The rCBV values showed highest correlation with total vessel area (MVA) (r=0.65, p=0.007) and slightly less correlation with vessel number (MVN) (r=0.52, p=0.04). Tumor showed significantly higher values than PTRE for all parameters: rCBV (1.87 + 0.82 versus 0.78 + 0.2, p=.002); MVA (0.22 + 0.03 versus 0.04 + 0.007, p=0.0001); and MVN (0.0068 + 0.001 versus. 0.0016 + 0.0006, p=0.0004). Conclusion: These preliminary results show the promise of Perfusion MRI to non-invasively estimate tissue microvessel density and distinguish tumor recurrence from treatment effects. The current study is ongoing to confirm results in a larger patient population. Citation Format: {Authors}. {Abstract title} [abstract]. In: Proceedings of the 101st Annual Meeting of the American Association for Cancer Research; 2010 Apr 17-21; Washington, DC. Philadelphia (PA): AACR; Cancer Res 2010;70(8 Suppl):Abstract nr 3748.