LaFond, Gabe

Background/Objectives: White matter (WM) tract detection is critical in the presurgical planning of tumor resection. However, standard-of-care imaging techniques including T1-weighted, T2-weighted, and Diffusion Tensor Imaging (DTI) often fail to identify WM tracts within edematous regions. In T1/T2-weighted imaging, edema increases extracellular water and reduces tissue contrast, and in diffusion-weighted imaging, edema elevates isotropic diffusion, reducing sensitivity to anisotropic diffusion along WM tracts. Advanced biophysical diffusion modeling techniques such as Neurite Orientation Dispersion and Density Imaging (NODDI) and the Standard Model (SM) address this limitation by compartmentalizing the diffusion signal into free-water, intra-neurite, and extra-neurite contributions. Here, we test if biophysical multi-compartment models can robustly identify WM tracts and recover tractography streamlines within edematous regions. Methods: In this study, we use multi-shell diffusion-weighted MRI data obtained from patients with meningiomas—a pathology allowing for isolation of the effects of edema without the confounding effects of tumor cell invasion. We compared FA from standard and free-water-corrected DTI, the orientation dispersion index (ODI) from NODDI, and P2 (a scalar descriptor of fiber orientation coherence) from the SM fODF in edematous and unaffected contralateral WM regions. As a proof of concept, we visually evaluated the tractography performance across models. Results: Our results show that (1 − ODI) and P2 values in edema remained close to within-subject contralateral measurements, contrasting with substantial reductions in FA and FW-FA. (1 − ODI) showed a small but statistically significant increase in edema (~8%, p = 0.02), while P2 was unchanged. Conclusions: These results highlight the potential of biophysical diffusion models for preoperative mapping in edema.

Abstract High-grade gliomas (HGG) have significant spatial heterogeneity, complicating therapeutic approaches. While multiparametric MRI (mpMRI) enables noninvasive assessment of this heterogeneity, most predictive models lack biological interpretability when used to classify or predict outcomes. Here, we developed an unsupervised framework that identifies novel mpMRI feature patterns through dimensional reduction, linking them to clinically relevant molecular signatures. We analyzed 18 advanced mpMRI features from 73 stereotactic biopsies across 20 HGG patients. Imaging data were normalized, followed by dimensional reduction using an iterative UMAP across varying parameters. Density-based clustering was used to generate stable image-derived clusters through consensus co-clustering. RNA-seq data from matched biopsies were analyzed using DESeq2 and GSEA to identify differentially expressed genes and enriched pathways. We then constructed a network linking specific imaging features to these transcriptional signatures. Validation was performed in an independent cohort, allowing us to extend clusters across full brain mpMRI. Unsupervised clustering of mpMRI features produced nine stable image-based clusters. These clusters were not fully explained by contrast enhancement, metabolic class, or tumor purity, shown by intraclass correlation (ICC) analyses, suggesting they capture novel biological variation. Transcriptomic analysis showed each cluster was associated with distinct gene expression patterns and enriched biological pathways, including neuronal differentiation, immune chemotaxis, and chromatin remodeling. Key imaging features (sRCBV, MTT, K2)—combined in cluster-specific patterns—enabled construction of an interpretable imaging–transcriptomic network. This was validated in an independent cohort. We then applied to full brain mpMRI, mapping spatially distinct regions to biological trends and patient outcomes. Addressing a research gap in imaging HGG, we present an interpretable network that connects specific patterns of mpMRI to clinically meaningful molecular signatures. This approach advances our understanding of tumor heterogeneity and supports future translational work in precision patient care.

Abstract BACKGROUND Glioblastoma (GBM) is a molecularly and phenotypically heterogeneous tumor, with regional invasion patterns driving treatment resistance and universal recurrence. In infiltrative regions, GBM cells adopt at least two predominant invasive phenotypes: Neuronal-enriched (NEU), with synaptic and neurodevelopment programs, and Glycolytic/Plurimetabolic (GPM), with heightened immune and metabolic activity. These subtypes differ in therapeutic vulnerability and microenvironmental interactions; features not captured by conventional contrast-enhanced (T1+C) MRI due to minimal blood-brain barrier (BBB) disruption. Our group sought to develop predictive models to distinguish variability of NEU/GPM subtypes within infiltrative GBM, utilizing graph convolutional networks (GCN), advanced multiparametric MRI (mpMRI), and spatially-matched image-localized biopsies with transcriptomic profiling. METHODS We analyzed 18 mpMRI features —including conventional, diffusion-weighted, and Dynamic Susceptibility Contrast (DSC) perfusion MRI— of 101 image-localized biopsies from MRI non-enhancing infiltrative margins of 64 GBM patient tumors. Features were hierarchically clustered per modality; each cluster was represented as a node in a Neo4j graph. Edges connected nodes sharing overlapping samples. Nodes were annotated with NEU or GPM subtype proportion. GCNs were trained to classify subtype-enriched nodes using graph topology and node features. RESULTS GCNs achieved high predictive performance on held-out test data (AUC = 0.95 for NEU, 0.88 for GPM; 70:30 train-test split) with similar 5-fold cross-validation performance and generalized well to an independent cohort (AUC > 0.85, n = 45). Importantly, performance was robust without relying on T1+C imaging, enabling prediction beyond BBB-disrupted regions. GCNs outperformed traditional classifiers (logistic regression, random forest). CONCLUSION Our results demonstrate that GCN-modeling of mpMRI features enables non-invasive prediction of invasive GBM subtypes at the tumor margin. Unlike convolutional neural networks (CNN), which rely on spatial continuity, GCNs leverage relational structure between heterogeneous imaging features, making them ideal for modeling sparse, non-grid data. This may support non-invasive regional vulnerability detection, surgical targeting, and improved response assessment.