Ranjbar, Sara

Abstract Magnetic resonance imaging (MRI) is a centerpiece of clinical management of brain tumor patients, yet the biology underlying even commonly interpreted imaging changes is unclear. The purpose of this project was to understand what biology is associated with brain tumor edema abundance and if that association was sex distinct. We have developed a two-compartment model of MRI signal intensity to quantitatively estimate relative edema abundance from T2-weighted MRIs. Multiparametric MRIs were preprocessed with bias field correction and intensity normalization. The images were co-registered and segmented for abnormal regions and normal tissue. The segmentations and preprocessed images were utilized in the edema mathematical model to estimate the fraction of extracellular space (Fecs) and fraction of intracellular space (Fics). Through an ongoing image-localized biopsy collection protocol, we analyzed 179 samples (Female: 75; Male: 104) from 55 patients (Female: 21; Male: 34) for bulk RNA-Seq. Samples were characterized by their edema Fecs scores and underwent differential expression, gene set enrichment analysis using MSigDB hallmarks and cellular deconvolution using CIBERSORTx. Through differential expression and gene set enrichment analyses of these spatial biopsies, we have identified sex-distinct gene expression and pathways corresponding to regions of high vs low edema. Within highly edematous regions, male samples were amplified for respiratory pathways, while female samples were suppressed. Connecting changes on MRIs with molecular markers from image-localized biopsies provides an opportunity to identify the biological drivers of brain tumor associated imageable edema. Specifically, elucidating the sex-distinct patterns connecting changes on MRI with cellular populations and molecular pathways could be used to better interpret imaging of patients and interpreting treatment response. Citation Format: Pamela R. Jackson, Lee Curtin, Sara Ranjbar, Kyle W. Singleton, Maciej M. Mrugala, Richard S. Zimmerman, Bernard R. Bendok, Peter Canoll, Kristin R. Swanson. Sex-distinct patterns of molecular pathways associated with brain tumor edema [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2024; Part 1 (Regular Abstracts); 2024 Apr 5-10; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2024;84(6_Suppl):Abstract nr 6168.

Abstract Identification of key phenotypic regions such as necrosis, contrast enhancement, and edema on magnetic resonance imaging (MRI) is important for understanding disease evolution and treatment response in patients with glioma. Manual delineation is time intensive and not feasible for a clinical workflow. Automating phenotypic region segmentation overcomes many issues with manual segmentation, however, current glioma segmentation datasets focus on pre-treatment, diagnostic scans, where treat-ment effects and surgical cavities are not present. Thus, existing automatic segmenta-tion models are not applicable to post-treatment imaging that is used for longitudinal evaluation of care. Here, we present a comparison of three-dimensional convolutional neural networks (nnU-Net architecture) trained on large temporally defined pre-treatment, post-treatment, and mixed cohorts. We used a total of 1563 imaging timepoints from 854 patients curated from 13 different institutions as well as diverse public data sets to understand the capabilities and limitations of automatic segmenta-tion on glioma images with different phenotypic and treatment appearance. We as-sessed the performance of models using Dice coefficients on test cases from each group comparing predictions with manual segmentations generated by trained techni-cians. We demonstrate that training a combined model can be as effective as models trained on just one temporal group. The results highlight the importance of a diverse training set, that includes images from the course of disease and with effects from treatment, in the creation of a model that can accurately segment glioma MRIs at multiple treatment time points.

Abstract Image-based mathematical modeling is emerging as an advantageous tool to integrate into the care routine for patients with brain tumors. These tools can be used for a variety of tasks including predicting tumor growth, indicating spatial genomic alterations, and guiding radiation therapy. Often, the methods to produce these maps are developed in silos, using cohorts of retrospective patients and switching to a per patient map calculation can be an afterthought. Automatic pipelines are needed to enable creating mathematical maps in a clinically-relevant workflow. To this end, we developed a pipeline to create an image-based mathematical map of edema abundance based on T2-weighted (T2W) magnetic resonance imaging that includes preprocessing steps, abnormality and tissue type segmentation, brain extraction, map generation, and conversion between image file formats. The python-based pipeline was designed to take in T1-weighted (T1W), T1W with gadolinium contrast (T1Gd), T2W, and fluid attenuated inversion recovery (FLAIR) DICOM images for a single patient as input and convert them to NIfTIs for downstream steps. The pipeline preprocessed images by correcting bias field fluctuations using the N4 algorithm and normalizing intensities. The T1W, T1Gd, and FLAIR are then registered to the T2W. T1Gd and FLAIR are used in a deep learning algorithm to delineate the brain and abnormality including enhancing tumor, necrosis, and peripheral edema. Normal tissue segmentations are performed using the statistical parametric mapping (SPM) segmentation routine. The segmentations and preprocessed images are all utilized in the edema mathematical model. The resulting edema map is then reformatted as a DICOM and output for further use by clinicians and clinical imaging systems. Our future goals include integrating our pipeline in a clinical workstation and sending final maps to our hospital’s picture archiving and communication system (PACS) to allow for deployment of these models for patient care in a clinically efficient timeline.

Abstract Brain tumor associated edema is a significant cause of patient morbidity and mortality often requiring continual treatment with differing levels of success across patients. Edema is a broad term that can indicate increased local water as well as diffusely infiltrating tumor cells. Edema is visualized as hyperintense regions on T2-weighted (T2W) magnetic resonance imaging (MRI) and is typically associated with blood-brain barrier (BBB) breakdown and tumor aggressiveness. Understanding the molecular mechanisms driving imaging patterns of “edema” could provide insights into clinical imaging interpretation. We have an ongoing image-guided biopsy study that allows us to link biopsy molecular markers with locoregional MRI patterns of edema. Further, we previously developed a physics-based method to estimate edema abundance (i.e., edema map) from T2W MRIs. Our goal was to identify connections between BBB-associated molecular factors and edema abundance in brain tumors. Our cohort included 38 patients (female: 15, male: 23) with 129 image-guided biopsies (female: 62, male: 67). We correlated image-localized edema map values with the mean transcriptional frequency for 57 genes related to BBB function. Additionally, we examined correlations separately according to patient reported sex (i.e., male and female) and imaging phenotype (i.e., ENH: enhancing and NE: non-enhancing). We utilized multiple comparisons corrections with a 5% false discovery rate to determine significance. For the overall cohort, we observed significant positive correlations for the HIF1A (p< 0.001) and SOX2 markers with edema. For NE samples, significant correlations included APOE (p=0.001), HIF1A (p< 0.001), PIK3CA (p< 0.001), PTCH1 (p< 0.001), and SOX2 (p< 0.001). Amongst female samples, a significant correlation with PTCH1 (p=0.002) was observed. There were no significant correlations noted for male and enhancing sub-cohorts. Significant correlations between molecular markers of BBB and edema map values could lead to clinical biomarkers for edema or tumor aggressiveness.

Abstract Glioblastoma (GBM) is the most aggressive primary brain tumor with a median survival of 14 months. GBMs are challenging to treat due to their heterogeneous nature. It has also been seen that these tumors have sex differences in their cellular subtypes as well as imaging. Radiomics has the potential to provide a non-invasive, spatial understanding of genetic and epigenetic diversity in these complex tumors and to aid in treatment planning. We have an ongoing study to obtain image-localized biopsies from GBM patients, allowing us to complete radiomic analysis and make connections between immunohistochemistry (IHC) and magnetic resonance imaging (MRI) features. We sought to determine if the patterns on imaging were correlated with underlying tumor biology. We focused on immunohistochemistry (IHC) markers of key features of tumor biology including SOX2 for stem-like tumor cells, CD68 for immune response and Ki67 for proliferation kinetics. Our study included 38 patients with a total of 99 biopsies (bxs): 27 males with 77 bxs and 11 females with 22 bxs. Biopsies were sectioned and stained for the SOX2, CD68, and KI67 markers. We computed 18 first-order radiomic features at each biopsy location for patients’ multimodal MRIs: T1W, T1Gd, T2W, FLAIR, apparent diffusion coefficient, diffusion weighted imaging (DWI) and susceptibility weighted imaging. We then performed correlation analysis between each radiomic feature and marker abundance for each IHC stain. Overall, we found sex-distinct patterns connecting imaging with these IHC markers. For example, amongst female patients, DWI held more prominent correlations with SOX2 than in males. Whereas there were more correlations between CD68 IHC abundance and T1Gd imaging features in males compared to females. Taken together, the overall patterns connecting locoregional imaging features to these IHC markers showed sex-distinct patterns suggesting the potential for sex to be an important biological variable when interpreting the biology underlying imaging changes.

Abstract Purpose Accurately predicting the clinical breast cancer subtypes could be extremely helpful for radiologists, pathologists, surgeons, and clinicians and inform future treatment prediction algorithms. Therefore, we evaluate and compare the accuracy of radiomic features extracted from contrast enhanced mammography (CEM) and magnetic resonance imaging (MRI) scans to make predictions to subtypes of breast cancer. Methods This HIPAA-compliant prospective single institution study was approved by the local institutional review board with written informed consent. Women with breast tumors 2 cm or larger underwent dynamic contrast-enhanced MRI and/or CEM for surgical staging. Semi-manual regions of interest were drawn by radiologist using Cancer Imaging Phenomics Toolkit (CaPTk). Radiomic features were obtained using PyRadiomics and MR- and CEM- based classification models were built on a low-dimensional representation of the features obtained via kernel principal component analysis. We subscribed to an ensemble tree-based learning approach called extremely randomized trees (ERT) to predict cancer subtypes captured via immunohistochemistry markers. Results For MRI analysis, 124 women with newly diagnosed breast cancer were included in the analysis comprising 49 HR+HER2-, 37 HR+HER2+, 11 HR-HER2+, and 27 HR-HER2- cases. For CEM analysis, models were built using data from 170 female patients including 74 HR+HER2-, 41 HR+HER2+, 14 HR-HER2+, and 43 HR-HER2-. CEM based model resulted in accuracies of 55%, 72%, 88%, and 71% respectively for HR+HER2-, HR+HER2+, HR-HER2+, and HR-HER2- whereas MRI based model alone led to accuracies of 54%, 62%, 89%, and 76% respectively for HR+HER2-, HR+HER2+, HR-HER2+, and HR-HER2-. Conclusions Radiomic features extracted from CEM and MR were strong predictors of breast cancer subtypes with CEM-based radiomic features performing slightly better, though not statistically significantly better (p = 0.82), than its MRI counterpart.

Abstract INTRODUCTION Glioblastoma (GBM) is a diffusely invasive primary brain tumor with significant spread of tumor cells to the periphery of visible image abnormality. Enhancement of Gadolinium (Gd) contrast agent on magnetic resonance imaging (MRI) has historically been considered a confirmation of local breakdown of the blood brain barrier (BBB) and sufficient drug delivery to the bulk of tumors. In this work, we used GBM-derived xenograft (PDX) models to compare drug delivery in GBM brain for high and low BBB-permeable drugs. MATERIALS AND METHODS Five patient-derived orthotopic xenograft models from two GBM cell lines (GBM39 and GBM12) were co-dosed with erlotinib and osimertinib, two drugs with low and high BBB-permeability, respectively. T1Gd and T2-weighted MRIs were acquired from all animals prior to model sacrifice. Tumors were manually segmented on denoised and standardized MRIs and intensity patterns were captured using first and second order statistical features in the moving 3x3 kernel. We compared drug levels found in Matrix Assisted Laser Desorption Ionization (MALDI) in T1Gd enhancement, T2 enhancement, and normal brain. We also performed linear regression modeling to predict drug levels using MRI features. Model performance was measured using root mean squared error (RMSE). RESULTS Our analysis showed correlations between imaging features and MALDI drug levels. Osimertinib had a uniform distribution across the brain for all animals and all cell lines, consistent with our expectation for a high BBB-penetrant drug. Erlotinib showed the highest drug levels in T2 for GBM39 and in T1Gd for GBM12. Regression models showed promising results for predicting Erlotinib with a low RMSE of 0.037. CONCLUSION Our preliminary results suggest MRI can be predictive of drug levels for low-BBB penetrant drugs. Understanding the relationship between MRIs and drug distribution in diffuse tumors can be beneficial to developing effective treatment.

Abstract Glioblastoma (GBM) is a devastating primary brain tumor known for its heterogeneity with a median survival of 15 months. Clinical imaging remains the primary modality to assess brain tumor response, but it is nearly impossible to distinguish between tumor growth and treatment response. Ki67 is a marker of active cell proliferation that shows inter- and intra-patient heterogeneity and should change under many therapies. In this work, we assessed the utility of a semi-supervised deep learning approach for regionally predicting high-vs-low Ki67 in GBM patients based on MRI. We used both labeled and unlabeled datasets to train the model. Labeled data included 114 MRI-localized biopsies from 43 unique GBM patients with available immunohistochemistry Ki67 labels. Unlabeled data included nine repeat routine pretreatment paired scans of newly-diagnosed GBM patients acquired within three days. Data augmentation techniques were utilized to enhance the size of our data and increase generalizability. Data was split between training, validation, and testing sets using 65-15-20 percent ratios. Model inputs were 16x16x3 patches around biopsies on T1Gd and T2 MRIs for labeled data, and around randomly selected patches inside the T2 abnormal region for unlabeled data. The network was a 4-conv layered VGG-inspired architecture. Training objective was accurate prediction of Ki67 in labeled patches and consistency in predictions across repeat unlabeled patches. We measured final model accuracy on held-out test samples. Our promising preliminary results suggest potential for deep learning in deconvolving the spatial heterogeneity of proliferative GBM subpopulations. If successful, this model can provide a non-invasive readout of cell proliferation and reveal the effectiveness of a given cytotoxic therapy dynamically during the patient's routine follow up. Further, the spatial resolution of our approach provides insights into the intra-tumoral heterogeneity of response which can be related to heterogeneity in localization of therapies (e.g. radiation therapy, drug dose heterogeneity).

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.