Swanson, Kristin R.

Abstract There have been increasing efforts to standardize MRI scanning protocols for both routine and clinical trial imaging of glioma. The recent BTIP guidelines propose acquisition of T2-weighted images after gadiolinium (Gd) based contrast injection and before post-contrast 3D T1-weighted (T1W) images to control timing of images after contrast administration. While Gd-contrast shortens T1W signal and appears hyperintense on magnetic resonance imaging (MRI), Gd-contrast also shortens T2W signal, which could result in unintended alternations in T2W signal intensities. This has potential implications for model-based approaches for GBM diagnosis and the differential diagnosis of other tumor types (e.g., lymphoma) that rely on T2W signal characteristics. To our knowledge, no neuroimaging studies have formally evaluated the impact of Gd contrast on T2W imaging. We acquired multi-echo T2 sequences (TE=14, 28, 41, 55, 69 ms) both prior to and after the administration of Gd for 5 patients suspected of having a brain tumor. Quantitative T2 maps were created by fitting the signal decay to a signal exponential with the Levenburg-Marqurdt algorithm. Percent difference maps were calculated to compare the quantitative T2 changes between the pre- and post-Gd T2 maps. Additionally, we performed a Wilcoxon Rank-Sum test comparing T2 values within regions of interest (ROIs) representing the tumor and normal appearing white matter (NAWM). Significant differences between the T2 values in the tumor and NAWM ROIs were seen for 4 and 3 of the patients, respectively (all p-values < 0.01). Our initial data suggests that the administration of Gd contrast may affect T2 values in both the tumor and NAWM, which could impact quantitative image-based models and diagnoses in glioma. We are currently replicating these preliminary results in an expanded cohort.

Abstract Determining effective therapies for patients with recurrent glioblastoma remains difficult especially in small early phase clinical trials. Currently utilized estimators of treatment response rely on two-dimensional measurements of tumor progression (e.g. RANO, RECIST, MacDonald) that have often fallen short in reliably connecting patients who respond with positive survival outcomes. This has prompted consideration of advanced imaging and volumetric measurements by the RANO working group and others. Our proposed approach, Days Gained (DG), uses volumetric measurements to calculate a personalized response metric based on change in tumor growth characteristics pre/post-treatment. DG has shown promise for connecting imageable treatment response with clinically meaningful (statistically significant) survival benefit when applied in upfront and recurrent therapy settings, even in the context of coincident bevacizumab treatment. In this study, we seek to consider a new cohort of patients receiving bevacizumab in combination with other novel therapies to distinguish any subpopulation of patients with meaningful response. In this work, we identified a subset of 16 patients from bevacizumab clinical trials with data available for DG analysis requiring serial imaging (T1Gd and FLAIR) prior to and after the first cycle of therapy. Each trial evaluated bevacizumab alone versus bevacizumab plus either dasatinib or TRC105 in recurrent disease. Discrimination of time to progression (TTP) was evaluated at DG thresholds identified from our prior analysis of patients treated with bevacizumab. Using previously identified DG thresholds for clinically significant treatment benefit, in our cohort, DG measured by serial FLAIR imaging was significant in predicting longer TTP (120 DG, p=0.014), but T1Gd DG scores did not attain significance (p=0.630). Given the limited insights in connecting response metrics to accurately predicted outcomes, these data further support the use of DG as a burgeoning marker of treatment response (or lack of response) that may be useful for more routine integration into clinical trials.

Abstract BACKGROUND Conventional magnetic resonance imaging (MR) guides patient care in GBM. However, there is mounting awareness that MR enhancement is non-specific reflecting either tumor progression or non-tumoral inflammatory changes. Histological evaluation of GBM is held as the gold standard for disease assessment. However, the invasiveness of this methodology and the sample sparsity limit its usefulness. Methods to infer histological underpinnings of MRI are needed to improve clinical care. METHODS A transfer learning mixed effects model based on T1Gd and FLAIR MR voxel based image features trained and cross-validated to predict FPKM values of CD68, CASP3, CD8A and IL13RA2. Training data included RNAseq from 38 image-localized biopsies from 15 newly-diagnosed GBM patients. These models were then applied to two independent patients, chosen based on therapy and the quality of image registration, at three different time points, just prior to receiving IL13RA2 targeted CAR-T therapy for rGBM, after 2 cycles and after 4 cycles, resulting in a voxel-based prediction of the relative expression levels of these genes. RESULTS Cross-validation of the proposed machine learning models demonstrated a Pearson Correlation coefficient between predicted and observed FPKM values of 0.89 (CD68), 0.92 (CASP3), 0.93 (CD8A), and 0.88 (IL13RA2). When the models were applied to the two CAR-T patients, CD68-modeled expression levels were seen to increase throughout therapy for one patient, while remaining stable for the other. Survival from first CAR-T infusion was, respectively, 412 and 83 days suggesting that the increased CD68 was indicative of therapeutic effectivity. Further, the spatial activity predictions were consistent with expected therapeutic action as the correlation coefficient between CASP3 and the product of IL13R2 and CD8A expression was 0.81, suggesting the T-cells targeting IL13RA2 were inducing cell death. CONCLUSIONS Preliminary results with our model highlights the potential of spatially resolved radiomic maps to provide insight into regional therapeutic effectivity.

Abstract PURPOSE Brain tumor related epilepsy (BTE) is a major co-morbidity in patients with glioma. It is difficult to determine whether the use of anti-epileptic drugs is necessary. We attempted to build a machine-learning model to predict the probability of seizure presentation (SP) with glioma. METHODS We trained a random forest classifier using the following variables: volumetric data of pre-treatment MR images (T1Gd and T2-FLAIR sequences), patient demographics (age; sex), and measurements of tumor proliferation (log(ρ)), invasiveness (log(D)) and their relative ratio (log(ρ/D)). Our cohort consisted of 221 patients total. Using an 80-20 ratio, we used 176 patients (76 SP, 100 nSP) for training and the remaining 45 patients (19 SP, 26 nSP) were used for testing. We also trained on male-only and female-only cohorts to evaluate any sex differences in prediction. For training, 108 males (53 SP, 55 nSP) were used and 28 for testing (14 SP, 14 nSP). We used 72 females (21 SP, 49 nSP) for training and 15 (7 SP, 8 nSP) for testing. We corrected for class imbalance in the female cohort before training. Using 10-fold cross-validation and a separate testing set, we measured performance by ROC curve (AUC), accuracy, sensitivity, and specificity of predictions (average of folds in cross validation). RESULTS The female model achieved the highest AUC (0.853) followed by the mixed model (0.726) and the male model (0.651). In the validation set, the accuracy/sensitivity/specificity of the three cohorts were as follows: mixed (0.726/0.696/0.750), female (0.853/0.830/0.875), and male (0.651/0.577/0.722). The performance of the testing set, in terms of accuracy/sensitivity/specificity were: mixed (0.733/0.74/0.73), female (0.8/0.57/1), and male (0.714/0.64/0.79). CONCLUSION We found a negative correlation between seizure probability and size and invasiveness of tumors. Our model shows promising performance on testing set data. Further cohort studies and training is warranted.

Abstract Recurrent glioblastoma is challenging to distinguish from so called “treatment effect” on routine clinical imaging. Further, within tumor heterogeneity reveals that some regions can be histologically dominated by tumor progression whilst others can be dominated by secondary effects of treatment response. Apparent tumor progression on MRI can be very difficult to manage clinically as it is unclear the degree to which the imaging changes are actually tumor progression vs response to treatment (including inflammatory response and necrosis). In this analysis, we study a unique cohort of patients for whom image localized-biopsies reveal heterogeneity in response vs progression. Our dataset included 70 biopsy samples from 32 patients with GBM each histolopatholgically characterized for tumor abundance vs immune infiltrate. Six multiparametric MRI contrasts were available, including T1, T1gd, T2, FLAIR, SWI, and ADC. Images were co-registered. Radiomic (statistical + texture) features were extracted from the region of six image contrasts locally matched with each biopsy sample. Machine learning models were built to predict each biomarker using radiomic features. Leave-one-out cross validation was used to evaluate the prediction accuracy. Radiomic features were found to be informative to the prediction of biomarkers. ANOVA tests show significant improvement of using radiomic features compared with the null model. The prediction accuracy was higher when considering the biomarkers on a binary scale using the median as the cutoff than on a numerical scale. Spatially-informed radiomics models for tumor progression vs treatment effect are possible and can play an instrumental role in navigating confounding imaging changed common during treatment progression.

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 High-grade gliomas (HGGs) nearly always recur after standard initial treatment, and the resulting mixture of recurrent tumor and treatment-induced reactive changes presents major diagnostic challenges. Anatomical imaging, such as MRI, cannot adequately distinguish progressive disease from treatment effect (pseudo-progression). Furthermore, there is marked intra-tumoral heterogeneity, such that some areas of a tumor may demonstrate necrotic treatment effect and others frank recurrence. Due to this difficulty reliably differentiating between these two clinical findings, analytic methods using multiple modalities are necessary to further our understanding of this disease process. To this end, we sought to correlate radiographic, histopathologic and molecular features of surgically sampled post-treatment suspected recurrence to identify markers distinguishing tumor growth from treatment effect. We performed Stimulated Raman Histology (SRH) imaging and highly multiplexed RNA-sequencing (PLATE-seq) on 84 MRI-localized biopsies from 39 patients with clinically suspected recurrent HGG. The SRH images were classified as recurrent tumor or gliotic/reactive tissue using a convolutional neural network trained on an independent cohort including a large set of recurrent HGG, and an automated cell-counting algorithm was used to quantify cellularity from the SRH image of each sample. Differential gene expression analysis of the PLATE-seq data was used to identify gene sets that distinguish recurrent tumor from treatment effect, and single sample gene set variation analysis (GSVA) was used to further assess the molecular and cellular composition of each MRI-localized sample. The histopathologic and molecular features of each sample were also correlated with the MRI features of the corresponding biopsy sites, and this data is currently being used to train machine learning models that predict the distribution of recurrent tumor and treatment-induced reactive changes within a patient’s radiographic lesion. These predictive radiomic models will help to guide neurosurgical sampling, and improve our ability to monitor glioma progression and response to therapy.

Abstract Glioblastomas (GBMs) are known to be complex tumors comprising multiple subpopulations of genetically-distinct cancer cells; it is thought that this genetic variation is a major factor in the lack of observed survival benefit of treatment regimes that target one of these subpopulations. The field of radiogenomics seeks to study correlations between MRI patterns and genetic features of GBM tumors. Spatial radiogenomic maps produced using machine-learning (ML) methods that are trained against information from image-localized patient biopsies identify regions where particular cancer sub-populations are predicted to occur within a GBM, thus non-invasively characterizing the regional genetic variability of these tumors. These tumor subpopulations may also interact with one another, in ways which may be of a competitive or cooperative nature to varying degrees. It is important to ascertain the nature of these interactions, as they may have implications for treatment response to targeted therapies, and characterization of the spatio-temporal dynamics of these co-evolving sub-populations will shed light on why some therapies fail. Here we combine mathematical modeling techniques and spatially-resolved radiogenomic maps to study the nature of these interactions between molecularly-distinct GBM subpopulations. We model the interactions between cell populations using a partial differential equation based formalism. The model is parameterized using radiogenomic ML maps from which we infer the nature of interactions between subpopulations. Furthermore, using maps as inputs, the model turns static maps into dynamic information, thus providing insight into how these subpopulations composing the tumor change over time and the effect this has on observed treatment response for individual patients.

Abstract INTRODUCTION Glioblastoma (GBM) is an aggressive primary brain tumor, known for its poor prognosis. Due to its diffuse invasiveness into normal-appearing brain, localized treatments such as surgical resection and radiotherapy are typically supplemented with chemotherapy. However, to reach invading tumor cells, such antineoplastic drugs must cross the blood-brain barrier (BBB). That is, while angiogenesis induces BBB breakdown in dense tumor regions, the BBB remains rather intact for invading GBM cells. As a result, it is unclear whether BBB-impermeable drugs are delivered at a sufficient level to be effective. METHODS In order to study heterogeneity in BBB breakdown, experiments were conducted using both flank and intracranial patient-derived xenografts (PDXs) treated with the EGFR-targeted monoclonal antibody drug conjugate, depatuxizumab mafodotin (ABT-414). Time-series bioluminescence imaging (BLI) data was used to develop a differential equation model of tumor growth for three PDX cell lines. Data from untreated PDXs, both in flank and intracranially, were used to parameterize tumor proliferation rates. Flank PDX data were used to parameterize individual sensitivity to ABT-414, whereas intracranial PDX data were used to determine the proportion of drug exposed to the tumor. RESULTS Each PDX line differed in response to the study drug ABT-414. As expected, such heterogeneous responses can primarily be attributed to differences in both drug sensitivity and the proportion of drug that reached the tumor. Notably, the estimated proportion of drug that reached the tumor was highest in the PDX line with the longest survival times, despite also having higher estimates of resistance. This suggests that PDXs with greater overall BBB breakdown may respond better to this agent. CONCLUSIONS Although more cell lines are needed to validate our approach, parameterizing this model for PDXs gives valuable insight into the extent of BBB breakdown in patient GBMs and may aid in determining optimal therapies for individual patients.