MRI-Based Radiogenomic Model Predicts Survival in Glioblastoma via IL-19

Researchers have built a machine-learning model that reads magnetic resonance imaging scans of glioblastoma tumors and infers the activity of a gene linked to poor survival, offering a non-invasive way to sort patients by expected outcome. The study, published in Cancer Imaging, focuses on interleukin-19, an immunosuppressive cytokine that dampens T-cell activation and has been associated with unfavorable prognosis in glioblastoma, the most aggressive form of adult glioma.

The team, led by Justin Bo-Kai Hsu and Gilbert Aaron Lee with senior author Cheng-Yu Chen, began by asking how the mixture of immune-signaling molecules inside a tumor shapes patient survival. Tumor-infiltrating immune cells secrete a range of immunoregulatory cytokines that collectively influence tumor progression, yet the combined prognostic impact of these signals in glioma had not been systematically characterized. To fill that gap, the researchers analyzed RNA-sequencing data from the TCGA glioma cohort, which covers World Health Organization grade II through IV tumors, and used consensus clustering to group patients according to cytokine-expression patterns.

That analysis revealed four molecular subgroups with significantly different survival outcomes. The subgroups with the poorest prognosis showed marked enrichment of M2-like macrophage signatures, a profile associated with immunosuppressive activity in the tumor microenvironment, and carried the highest levels of IL19 expression. Immune cell enrichment was evaluated by gene set enrichment analysis using curated Panglao-derived marker sets, allowing the investigators to tie the cytokine patterns to specific immune cell populations within the tumors.

Interleukin-19 emerged as a consistent marker of aggressive disease. The cytokine was strongly associated with adverse prognosis and showed significantly increased expression in grade IV gliomas across both the TCGA and CGGA datasets. Because IL-19 attenuates T-cell activation, the authors argue it highlights a mechanism by which the tumor microenvironment suppresses antitumor immunity, making the cytokine a potential therapeutic target as well as a prognostic one.

The second half of the study addressed a practical problem: IL19 expression can only be measured directly from tumor tissue, which requires biopsy or resection. The researchers therefore sought to predict it from imaging. They assembled histologically confirmed grade IV glioblastoma cases that had MRI data available through the Cancer Imaging Archive and matched transcriptomic data, both RNA-sequencing and microarray, from the TCGA cohort. Radiomic features were extracted from contrast-enhanced T1-weighted images and from apparent diffusion coefficient maps, which capture water diffusion within tissue.

Using Bayesian logistic regression models with backward feature selection, the team constructed predictors of IL19 expression status. Multiple radiomic features correlated with IL19 levels, and these correlations were found predominantly within first-order intensity classes, meaning that simple statistical descriptors of pixel intensity carried much of the predictive signal. The models achieved areas under the receiver operating characteristic curve of 0.84 to 0.85 and accuracies of 0.77 to 0.79 for predicting IL19 expression, and they were validated in independent GBM microarray and RNA-sequencing cohorts rather than only in the data used to train them.

The validation went beyond classification accuracy. When patients were grouped by the IL19 status predicted from their scans, the model-predicted IL19-high groups consistently exhibited significantly worse survival. This indicates that the imaging-derived estimate tracks not only the molecular signature but also the clinical consequence associated with it, effectively converting a routine multiparametric MRI examination into a surrogate readout of tumor immune biology.

The authors describe the approach as radiomic immunophenotyping, a framework that links imaging phenotypes to intratumoral immune expression without the need for additional tissue sampling. In their conclusions they position the model as an exploratory platform for investigating immune heterogeneity in glioblastoma, a disease in which immunotherapy has shown limited efficacy, partly because immunosuppressive mechanisms within the tumor microenvironment are highly heterogeneous. A tool that stratifies patients by cytokine-driven immune profiles could, in principle, help identify which individuals carry tumors dominated by macrophage-mediated suppression.

Several caveats frame the work. The clustering and prognostic associations come from retrospective analysis of existing cohorts, and the radiogenomic model was trained and tested on patients with matched imaging and transcriptomics, a subset that may not represent every clinical presentation. The study reports associations between IL19 expression, macrophage enrichment and survival; it does not demonstrate that blocking IL-19 improves outcomes, and the cytokine’s status as a therapeutic target remains to be tested in interventional studies. The article was shared as an accepted manuscript under a permanent DOI, with a final version of record to follow.

Even so, the findings add to a growing effort to extract molecular information from medical images. By showing that features as basic as intensity statistics on T1-weighted and ADC sequences can predict the expression of a specific immunosuppressive cytokine, and that the prediction separates patients by survival, the study suggests a route toward immune characterization of glioblastoma at the time of diagnosis, using scans that patients already undergo as part of standard care.

Subject of Research: A radiogenomics study linking interleukin-19 expression to glioblastoma survival and MRI-based prediction

Article Title: Interleukin-19 as a novel target for predicting survival in glioblastoma patients using a radiogenomics approach

Article References: Hsu, J. B.-K., Lee, G. A., Tsao, Y.-W., Kuo, D.-P., Lu, C.-F., Chen, Y.-C., & Chen, C.-Y. (2026). Interleukin-19 as a novel target for predicting survival in glioblastoma patients using a radiogenomics approach. Cancer Imaging. https://doi.org/10.1186/s40644-026-01151-4

Image Credits: AI Generated

DOI: 10.1186/s40644-026-01151-4

Keywords: glioblastoma, interleukin-19, radiogenomics, MRI, tumor microenvironment, immunosuppression, cytokines, machine learning, prognosis, TCGA, radiomics, macrophages