A single-centre study published on 10 October in Scientific Reports examined non-contrast 3T MRI scans from 41 patients whose emergency diagnoses raised a difficult distinction between low-grade glioma and encephalitis. Researchers at the University of Rome Tor Vergata and affiliated hospital units trained random-forest classifiers on radiomic features from T2-FLAIR, diffusion and T1 scans. The cohort was split into 31 training cases and 10 for independent validation. FLAIR-based models performed best; the model combining radiomics with clinical data across sequences had an AUC of 0.80.

For that combined-sequence model, the paper reports 100% sensitivity, 80% specificity and 90% accuracy on the validation set. But only 10 patients were in that set, and the confidence intervals were wide: the FLAIR AUC was 0.82 with a 95% interval of 0.54–1.00. The authors describe the work as proof of concept and say larger multicentre studies are needed before clinical use.