Biologically-informed deep neural networks provide quantitative assessment of intratumoral heterogeneity in post-treatment glioblastoma

Intratumoral heterogeneity poses a significant challenge to the diagnosis and treatment of glioblastoma (GBM). This heterogeneity is further exacerbated during GBM recurrence, as treatment-induced reactive changes produce additional intratumoral heterogeneity that is ambiguous to differentiate on clinical imaging. There is an urgent need to develop non-invasive approaches to map the heterogeneous landscape of histopathological alterations throughout the entire lesion for each patient. We propose to predictively fuse Magnetic Resonance Imaging (MRI) with the underlying intratumoral heterogeneity in recurrent GBM using machine learning (ML) by leveraging image-localized biopsies with their associated locoregional MRI features. To this end, we develop BioNet, a biologically-informed neural network model, to predict regional distributions of three tissue-specific gene modules: proliferating tumor, reactive/inflammatory cells, and infiltrated brain tissue. BioNet offers valuable insights into the integration of multiple implicit and qualitative biological domain knowledge, which are challenging to describe in mathematical formulations. BioNet performs significantly better than a range of existing methods on cross-validation and blind test datasets. Voxel-level prediction maps of the gene modules by BioNet help reveal intratumoral heterogeneity, which can improve surgical targeting of confirmatory biopsies and evaluation of neuro-oncological treatment effectiveness. The non-invasive nature of the approach can potentially facilitate regular monitoring of the gene modules over time, and making timely therapeutic adjustment. These results also highlight the emerging role of ML in precision medicine.

Re-convolving the compositional landscape of primary and recurrent glioblastoma reveals prognostic and targetable tissue states

Glioblastoma (GBM) diffusely infiltrates the brain and intermingles with non-neoplastic brain cells, including astrocytes, neurons and microglia/myeloid cells. This complex mixture of cell types forms the biological context for therapeutic response and tumor recurrence. We used single-nucleus RNA sequencing and spatial transcriptomics to determine the cellular composition and transcriptional states in primary and recurrent glioma and identified three compositional 'tissue-states' defined by cohabitation patterns between specific subpopulations of neoplastic and non-neoplastic brain cells. These tissue-states correlated with radiographic, histopathologic, and prognostic features and were enriched in distinct metabolic pathways. Fatty acid biosynthesis was enriched in the tissue-state defined by the cohabitation of astrocyte-like/mesenchymal glioma cells, reactive astrocytes, and macrophages, and was associated with recurrent GBM and shorter survival. Treating acute slices of GBM with a fatty acid synthesis inhibitor depleted the transcriptional signature of this pernicious tissue-state. These findings point to therapies that target interdependencies in the GBM microenvironment.

OTEH-6. Algorithmic approach to characterize post-treatment recurrent glioma using RNA sequencing and quantitative histopathology

Introduction

Distinguishing between tumor and treatment effect in post-treatment glioma, although crucial for clinical management, is difficult because contrast-enhancing regions are mixtures of recurrent tumor and reactive tissue, and definitive histopathological criteria do not exist. This study disentangles the marked intra-tumoral heterogeneity in the treatment-recurrent setting by developing an unsupervised framework to algorithmically categorize intraoperative MRI-localized biopsies into three clinically-relevant tissue clusters based on joint analysis of RNA sequencing and histopathological data.

Methods

A retrospective cohort of 84 MRI-localized biopsies from 37 patients with post-treatment recurrent glioblastoma underwent mRNA extraction and quantification via PLATEseq protocol. For 48 of 84 biopsies, a neighboring piece of tissue underwent quantitative histopathology based on labeling index (LI) for SOX2, CD68, NeuN, Ki67, and H&E. Correlation between LIs and gene expression for these 48 samples was performed. Genes significantly correlated (p<0.05) with ≥1 marker were used for hierarchical clustering of correlation matrix, identifying three mutually-exclusive tissue-specific gene sets. These sets were then used to perform ssGSEA to categorize each of 84 biopsies into one of three tissue types.

Results

Correlation analysis identified 7779 genes significantly correlated with ≥1 histopathological marker. Clustering revealed three gene sets associated with specific markers: SetA-3688 genes associated with SOX2/Ki67/H&E; SetB-2418 genes associated with CD68; SetC-1673 genes associated with NeuN. ssGSEA using these sets categorized each biopsy into one of three tissue types: 27 biopsies enriched in SetA, 28 in SetB, and 29 in SetC.

Conclusions

Using MRI-localized biopsies with both RNAseq and histopathological data, this algorithmic approach allows development of three orthogonal tissue-specific gene sets that can be applied to characterize the heterogeneity in post-treatment recurrent glioma: SetA: genes correlated with SOX2/Ki67/H&E, representing recurrent tumor; SetB: genes correlated with CD68 (microglia) representing reactive tissue consistent with treatment effect; SetC: genes correlated with NeuN (neurons), representing infiltrated brain.

NIMG-69. RAPID INTRAOPERATIVE DIAGNOSIS OF GLIOMA RECURRENCE USING STIMULATED RAMAN HISTOLOGY AND DEEP NEURAL NETWORKS

Accurate intraoperative diagnosis of recurrence versus treatment effect (TE) is essential for determining the management of suspected recurrent gliomas. Cytologic and histoarchitectural changes related to chemoradiation overlap with common findings in recurrent tumors (e.g. atypia, abnormal vasculature, necrosis). Moreover, H&E tissue processing artifact complicates interpretation. Stimulated Raman histology (SRH) uses the intrinsic biochemical properties of fresh, unprocessed surgical specimens to provide rapid label-free digital histologic images. Here, we report an automated technique using deep convolutional neural networks (ConvNet) that differentiates recurrent glioma and TE in fresh surgical specimens imaged using SRH with equivalent accuracy and 10x faster (tissue-to-diagnosis, 2 minutes) than conventional methods. Our ConvNet, based on Google’s Inception-ResNet-v2 architecture, was first trained on 3.6 million SRH images from 441 patients with the most common brain tumor subtypes. To optimize the network for classifying glioma recurrence, we used cross-validation (CV) on 35 patients (24 recurrent, 9 TE) for model hyperparameter tuning and to identify an optimal probability threshold to classify recurrence. To perform rigorous model validation, we used a 50 patient external testing set to evaluate overall model accuracy. Over 5 iterations of CV, the mean held-out classification accuracy was 94.8% (range, 91.4 - 97.1%). Using ROC analysis, we found that a probability of recurrence greater than 25% was the optimal threshold to render a recurrence diagnosis for whole-slide SRH images. Using our external testing set, we achieved a classification accuracy of 96% (total 48/50; 30/30 recurrences, 18/20 TE). Moreover, our method effectively identifies regions of glioma recurrence in whole slide SRH at no additional computational cost. Our study demonstrates the feasibility of applying deep learning for intraoperative diagnosis of recurrent gliomas in SRH imaged tissues. In the future, ConvNets may ultimately be used to guide decision-making in the surgical care of recurrent gliomas, independent of conventional neuropathology resources.

TMOD-14. RADIOGRAPHIC, STIMULATED RAMAN HISTOLOGIC, AND MULTIPLEXED RNA-SEQUENCING ANALYSIS OF POST-TREATMENT RECURRENT HIGH-GRADE GLIOMAS

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.

PATH-57. MRI-LOCALIZED BIOPSIES REVEAL HISTOPATHOLOGIC HETEROGENEITY IN POST-TREATMENT RECURRENT HIGH-GRADE GLIOMA

Evaluation of recurrence in post-treatment glioma is challenging because contrast-enhancing (CE) lesions are a mixture of tumor and treatment effect. This study characterizes intratumoral heterogeneity using quantitative digital pathology to correlate intraoperative MRI-localized biopsies with histopathology in the post-treatment setting. Findings will inform multiparametric radiographic models of intratumoral heterogeneity. A retrospective review was performed on adult patients with MRI-localized biopsies obtained during resection for post-treatment recurrent high-grade glioma. 68 patients and 170 MRI-localized samples were analyzed (median 2 samples/patient). Immunohistochemistry (IHC) for markers of glioma cells (SOX2), macrophages (CD68), and proliferating cells (KI67) was used to characterize biopsies. Slides were digitized and quantified using an automated cell-counting algorithm. Histopathological criteria based on IHC data was developed to classify biopsies. IHC quantification was compared across histological groups using ANOVA and paired t-tests. Most patients (52/68) yielded multiple biopsies. 75% (39/52) demonstrated heterogeneity in histological classification of all specimens obtained from their lesion. 47/170 (28%) biopsies were predominantly treatment effect, and most were CE (31/47 or 66%). Only 75/170 (44%) biopsies contained recurrent glioma, and 21/75 (28%) were NE. SOX2 labeling index was higher in biopsies containing recurrent tumor (p=5.13E-25). CD68 labeling index was higher in biopsies with predominant treatment effect (p=1.35E-12). IHC data from MRI-localized biopsies informed a multiple linear regression model which demonstrated significant predictive value for determining the distribution of recurrent tumor in the post-treatment setting. Contrast enhancement is not a reliable predictor of tumor in recurrent high-grade glioma. Most patients demonstrated marked intratumoral heterogeneity, highlighting the difficulty of accurate tumor sampling post-treatment glioma. Our histopathological classification significantly distinguished recurrent tumor from treatment effect and informed a multiparametric radiomic model which can guide surgical sampling and assess response to therapy.

NIMG-61. USING MACHINE LEARNING TO BUILD RADIOMICS MODELS THAT DISTINGUISH REGIONS OF GLIOBLASTOMA RECURRENCE VS TUMOR PROGRESSION ON MRI

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.

RADI-14. FRAMELESS STEREOTACTIC RADIOSURGERY ON THE GAMMA KNIFE ICON: EARLY EXPERIENCE FROM 42 PATIENTS WITH BRAIN METASTASES

BACKGROUND: The Gamma Knife (GK) Icon uses a Cone-Beam CT (CBCT) scanner and an infrared camera system to support the delivery of frameless radiosurgery. There are limited data on patients treated with frameless GK radiosurgery (GKRS) for brain metastases. OBJECTIVE: To describe the early experience, process, technical details, and short-term outcomes with frameless GKRS for brain metastases at our institution. METHODS: We describe our patient selection and workflow for frameless GKRS in detail. Because of the short interval of follow-up, we provide crude rates of local control. RESULTS: 42 patients had a total of 96 brain metastases. Median age was 69. 77 intact lesions were treated definitively, 18 cavities postoperatively, and 1 had GKRS for recurrence after resection. 11 patients underwent repeat GKRS to the same area. Median dose was 20Gy in 1 fraction (range: 14–21), 24Gy in 3 fractions (range: 19.5–27), and 25Gy in 5 fractions (Range: 25–30). Median treatment time was 23.7 minutes (Range: 7.3 – 85.5). 29 patients had a follow-up MRI in our records after completing GKRS. Median follow-up time was 105 days (Range: 16 – 314). 16 local recurrences (LR) were identified in 9 patients. An additional 6 patients had distant brain recurrence without LR. Crude mean time between GKRS and LR was 101 days (range 44–161 days). There were 6 patients with grade 1, 3 with grade 2, 2 with grade 3, and 1 with grade 4 toxicity. We found an improvement in workflow and a greater number of patients eligible for GKRS due to the ability to fractionate treatments. CONCLUSION: We report a large cohort of consecutive patients with brain metastases treated with frameless GKRS. We look forward to studies with longer follow-up to provide valuable data on clinical outcomes and to further our understanding of the radiobiology of hypofractionation in the brain.