Faculty Bibliography

BACKGROUND/UNASSIGNED:The isocitrate dehydrogenase (IDH) mutation status is a key molecular marker in diagnosing and treating brain tumors. Currently, it is determined via invasive tissue biopsy. Recent advances in deep learning (DL) have offered promising non-invasive alternatives for determining IDH status. However, their clinical translation is hindered by a significant gap between DL predictions and their clinical applicability. The limited transparency of many DL-networks and inadequate evaluation metrics hinders trust and adoption, as clinicians require clear and validated insights for determining IDH status. These challenges highlight the need for robust validation and measures of predictive reliability to make DL-predictions clinically actionable. METHODS/UNASSIGNED:We developed a unique approach for non-invasive prediction of IDH status using MRI. We combine a voxel-wise-segmentation network(MC-net) with Bayesian logistic regression (BLR) to provide an IDH status and estimate confidence scores. We utilized a comprehensive dataset of 2,481 glioma cases from eight institutions. RESULTS/UNASSIGNED:Our framework(MC-net + BLR) demonstrated robust performance achieving 96.4% and 95.1% classification accuracies on diverse databases, with an AUC of 0.98. The BLR was implemented exclusively on held-out test data, ensuring that the derived confidence scores are independent of the training or validation phases. The derived confidence scores showed a low Brier score of 0.0125, highlighting its superior calibration and uncertainty quantification. CONCLUSION/UNASSIGNED:The developed framework provides an IDH status and a confidence score, offering clinicians an additional layer of assurance in prediction reliability. It bridges the gap between high-performing DL models and their clinical applicability by addressing the challenges in prediction reliability. Our framework is a significant advancement in non-invasive determination of IDH-status and confidence-informed therapeutic decision-making in neuro-oncology.

BACKGROUND AND OBJECTIVES/OBJECTIVE:Vestibular schwannomas demonstrate different responses after stereotactic radiosurgery (SRS), commonly including a transient loss of internal enhancement on postcontrast T1-weighted MRI thought to be due to an early reduction in tumor vascularity. We used dynamic contrast-enhanced based golden-angle radial sparse parallel (GRASP) MRI to characterize the vascular permeability changes underlying this phenomenon, with correlations to long-term tumor regression. METHODS:Consecutive patients with vestibular schwannoma who underwent SRS between 2017 and 2019, had a transient loss of enhancement after SRS, and had long-term longitudinal GRASP studies (6, 18, and 30 months) were included in this retrospective cohort analysis (n = 19). Using GRAVIS ( https://gravis-imaging.org/gravis/ ), an analysis pipeline for GRASP studies, we extracted the key parameters normalized to the venous sinus from a region of interest within the tumor. RESULTS:The peak, area under the curve (AUC), and wash-in phase slope were significantly reduced at 6, 18, and 30 months after SRS (corrected P < .05), even while the internal enhancement returned in the tumors. Larger pre-SRS tumors were more likely to have a greater reduction in peak ( P = .013) and AUC ( P = .029) at 6 months. In a subset of patients (N = 13) with long-term follow-up, the median percentage reduction in tumor volume was 58% at a median of 62 months. These patients showed a strong correlation between peak, AUC, and wash-in phase slope changes at 6 months and tumor volume at the last follow-up. CONCLUSION/CONCLUSIONS:After SRS and loss of internal contrast uptake within vestibular schwannomas, a slow vascular permeability dynamic persisted, suggesting the presence of postradiation processes such as fibrosis. We show for the first time, using GRASP, a quantitative assessment of the vascular radiobiological effect.

PURPOSE/UNASSIGNED:Glioblastoma (GBM) is the most common and aggressive primary adult brain tumor. The standard treatment approach is surgical resection to target the enhancing tumor mass, followed by adjuvant chemoradiotherapy. However, malignant cells often extend beyond the enhancing tumor boundaries and infiltrate the peritumoral edema. Traditional supervised machine learning techniques hold potential in predicting tumor infiltration extent but are hindered by the extensive resources needed to generate expertly delineated regions of interest (ROIs) for training models on tissue most and least likely to be infiltrated. APPROACH/UNASSIGNED:We developed a method combining expert knowledge and training-based data augmentation to automatically generate numerous training examples, enhancing the accuracy of our model for predicting tumor infiltration through predictive maps. Such maps can be used for targeted supra-total surgical resection and other therapies that might benefit from intensive yet well-targeted treatment of infiltrated tissue. We apply our method to preoperative multi-parametric magnetic resonance imaging (mpMRI) scans from a subset of 229 patients of a multi-institutional consortium (Radiomics Signatures for Precision Diagnostics) and test the model on subsequent scans with pathology-proven recurrence. RESULTS/UNASSIGNED:Leave-one-site-out cross-validation was used to train and evaluate the tumor infiltration prediction model using initial pre-surgical scans, comparing the generated prediction maps with follow-up mpMRI scans confirming recurrence through post-resection tissue analysis. Performance was measured by voxel-wised odds ratios (ORs) across six institutions: University of Pennsylvania (OR: 9.97), Ohio State University (OR: 14.03), Case Western Reserve University (OR: 8.13), New York University (OR: 16.43), Thomas Jefferson University (OR: 8.22), and Rio Hortega (OR: 19.48). CONCLUSIONS/UNASSIGNED:The proposed model demonstrates that mpMRI analysis using deep learning can predict infiltration in the peri-tumoral brain region for GBM patients without needing to train a model using expert ROI drawings. Results for each institution demonstrate the model's generalizability and reproducibility.

PURPOSE/OBJECTIVE:DNA methylation profiling stratifies isocitrate dehydrogenase (IDH)-mutant astrocytomas into methylation low-grade and high-grade groups. We investigated the utility of the T2-FLAIR mismatch sign for predicting DNA methylation grade and cyclin-dependent kinase inhibitor 2A/B (CDKN2A/B) homozygous deletion, a molecular biomarker for grade 4 IDH-mutant astrocytomas, according to the 2021 World Health Organization (WHO) classification. EXPERIMENTAL DESIGN/METHODS:Preoperative MRI scans of IDH-mutant astrocytomas subclassified by DNA methylation profiling (n=71) were independently evaluated by two radiologists for the T2-FLAIR mismatch sign. The diagnostic utility of T2-FLAIR mismatch in predicting methylation grade, CDKN2A/B status, copy number variation, and survival was analyzed. RESULTS:The T2-FLAIR mismatch sign was present in 21 of 45 (46.7%) methylation low-grade and 1 of 26 (3.9%) methylation high-grade cases (p<0.001), resulting in 96.2% specificity, 95.5% positive predictive value, and 51.0% negative predictive value for predicting low methylation grade. The T2-FLAIR mismatch sign was also significantly associated with intact CDKN2A/B status (p=0.028) with 87.5% specificity, 86.4% positive predictive value, and 42.9% negative predictive value. Overall multivariable Cox analysis showed that retained CDKN2A/B status remained significant for PFS (p=0.01). Multivariable Cox analysis of the histologic grade 3 subset, which was nearly evenly divided by CDKN2A/B status, CNV, and methylation grade, showed trends toward significance for DNA methylation grade with OS (p=0.045) and CDKN2A/B status with PFS (p=0.052). CONCLUSIONS:The T2-FLAIR mismatch sign is highly specific for low methylation grade and intact CDKN2A/B in IDH-mutant astrocytomas.

The most widely used fluorophore in glioma-resection surgery, 5-aminolevulinic acid (5-ALA), is thought to cause the selective accumulation of fluorescent protoporphyrin IX (PpIX) in tumour cells. Here we show that the clinical detection of PpIX can be improved via a microscope that performs paired stimulated Raman histology and two-photon excitation fluorescence microscopy (TPEF). We validated the technique in fresh tumour specimens from 115 patients with high-grade gliomas across four medical institutions. We found a weak negative correlation between tissue cellularity and the fluorescence intensity of PpIX across all imaged specimens. Semi-supervised clustering of the TPEF images revealed five distinct patterns of PpIX fluorescence, and spatial transcriptomic analyses of the imaged tissue showed that myeloid cells predominate in areas where PpIX accumulates in the intracellular space. Further analysis of external spatially resolved metabolomics, transcriptomics and RNA-sequencing datasets from glioblastoma specimens confirmed that myeloid cells preferentially accumulate and metabolize PpIX. Our findings question 5-ALA-induced fluorescence in glioma cells and show how 5-ALA and TPEF imaging can provide a window into the immune microenvironment of gliomas.

Isocitrate dehydrogenase (IDH) mutation status has emerged as an important prognostic marker in gliomas. This study sought to develop deep learning networks for non-invasive IDH classification using T2w MR images while comparing their performance to a multi-contrast network. Methods: Multi-contrast brain tumor MRI and genomic data were obtained from The Cancer Imaging Archive (TCIA) and The Erasmus Glioma Database (EGD). Two separate 2D networks were developed using nnU-Net, a T2w-image-only network (T2-net) and a multi-contrast network (MC-net). Each network was separately trained using TCIA (227 subjects) or TCIA + EGD data (683 subjects combined). The networks were trained to classify IDH mutation status and implement single-label tumor segmentation simultaneously. The trained networks were tested on over 1100 held-out datasets including 360 cases from UT Southwestern Medical Center, 136 cases from New York University, 175 cases from the University of Wisconsin-Madison, 456 cases from EGD (for the TCIA-trained network), and 495 cases from the University of California, San Francisco public database. A receiver operating characteristic curve (ROC) was drawn to calculate the AUC value to determine classifier performance. Results: T2-net trained on TCIA and TCIA + EGD datasets achieved an overall accuracy of 85.4% and 87.6% with AUCs of 0.86 and 0.89, respectively. MC-net trained on TCIA and TCIA + EGD datasets achieved an overall accuracy of 91.0% and 92.8% with AUCs of 0.94 and 0.96, respectively. We developed reliable, high-performing deep learning algorithms for IDH classification using both a T2-image-only and a multi-contrast approach. The networks were tested on more than 1100 subjects from diverse databases, making this the largest study on image-based IDH classification to date.