Reshaping glioma imaging via non-invasive non-enhanced MR imaging methods: the GLIOCARE project

SUMMARY Contrast-enhanced MRI using gadolinium-based contrast agents (GBCAs) is a standard approach in brain tumor imaging due to its utility in lesion visualization, tumor characterization, and treatment planning. However, GBCAs accumulate in the body, pose environmental risks, and contribute to healthcare costs. They have been detected in drinking water and rivers, with adverse ecological effects. Despite their widespread use, there is no definitive evidence proving GBCA-enhanced MRI is superior to GBCA-free imaging for gliomas. This thesis investigates whether GBCA-free MRI protocols can reliably answer clinical questions in adult-type diffuse gliomas while maintaining diagnostic quality. Chapter 2 presents a comprehensive non-systematic literature review assessing the current status of GBCA-free and reduced-dose imaging for gliomas and meningiomas. While most clinical guidelines still recommend routine use of GBCAs, some support GBCA-free strategies, such as in pediatric low-grade gliomas and small, asymptomatic meningiomas. Two prospective trials have shown that reducing GBCA doses to 50–75% does not significantly affect diagnostic accuracy. Furthermore, interest in advanced GBCA-free techniques, like arterial spin labeling (ASL), amide proton transfer (APT-CEST), MR spectroscopy (MRS), and artificial intelligence (AI)-based synthetic imaging, has grown. T2-FLAIR-guided approaches for glioblastoma surgery and radiation therapy have shown clinical benefits. These findings suggest that GBCA-free imaging is increasingly feasible in select clinical scenarios, though more evidence is needed for widespread adoption in diffuse gliomas. Chapter 3 focuses on the VASARI feature set, a standardized collection of 30 glioma imaging descriptors largely derived from GBCA-free MRI. Through a systematic review and meta-analysis of 35 studies including 3,304 patients, I found that VASARI features were commonly used to predict overall survival and IDH mutation status. Key features such as multifocality, ependymal invasion, and enhancement crossing midline were significant survival predictors. Models using VASARI features reached a pooled AUC of 0.82 for IDH mutation prediction. However, heterogeneity in feature usage and scoring across studies limited reproducibility, emphasizing the need for automatic feature extraction to improve clinical applicability. In Chapter 4, I developed the enhancement prediction decision tree (EPDT) to determine whether features typically visualized with contrast could be predicted from GBCA-free MRI. The EPDT incorporated four imaging features (three VASARI, one non-VASARI) and was applied to 303 glioma cases by three raters with varying experience. The model accurately predicted enhancement quality (86–92%) and shape (84–89%) and showed substantial agreement with actual contrast-enhanced features. These results indicate that visually assessing GBCA-free sequences can substitute for contrast-enhanced imaging in characterizing glioma enhancement. Chapter 5 presents a diagnosis prediction decision tree (DPDT) built from seven VASARI and four non-VASARI features to predict glioma grade and molecular markers (IDH and 1p/19q status). Using GBCA-free MRI, raters achieved ≥85% accuracy for tumor grading and ≥75% for molecular classification, comparable to results using GBCA-enhanced imaging. Inter-rater agreement was moderate to substantial, indicating that GBCA-free imaging can be a reliable alternative for preoperative glioma diagnosis. In Chapter 6, I evaluated visual versus region-of-interest (ROI)-based assessment of diffusion-weighted imaging (DWI), the most widely used GBCA-free advanced imaging technique. Although ROI-based methods showed better reproducibility, both approaches correlated strongly and performed similarly in IDH mutation prediction. Visual assessment reached 69% accuracy, compared to 70–73% for ROI-based methods. In summary, this thesis demonstrates that GBCA-free MRI can answer key diagnostic questions in adult-type diffuse gliomas when paired with structured visual analysis and decision tree models. These findings support the development of safer, more sustainable, and cost-effective imaging strategies in neuro-oncology.