Qualitative MRI Features for Differentiating GBM, Solitary Brain Metastasis, and Primary CNS Lymphoma: A Retrospective Diagnostic Accuracy Study.

Background: Glioblastoma (GBM), solitary brain metastasis (SM), and primary central nervous system lymphoma (PCNSL) are important malignant brain tumors that may exhibit overlapping imaging features, making preoperative differentiation challenging. Although conventional MRI (cMRI) is widely available and routinely used in clinical practice, comprehensive comparative evaluation of cMRI features for differentiating these three entities remains limited. Purpose: This study aimed to identify conventional MRI features distinguishing GBM, SM, and PCNSL, and to develop a diagnostic algorithm for initial evaluation. Material and methods: This retrospective study included 108 patients (59 GBM, 24 SM, 25 PCNSL) who underwent pre-operative 3T brain MRI between May 2021 and May 2025, with histopathologic confirmation. The imaging features were compared using appropriate statistical tests. Statistically significant qualitative variables were then entered into a classification model built with the CHAID (Chi-square Automatic Interaction Detection) growing method. Results: Significant differences were observed across tumor types in lesion location, border characteristics, enhancement patterns, and specific signs such as the notch sign, worm-like sign, and intra-tumoral vascular enhancement (all p < 0.05). PCNSLs commonly showed homogeneous enhancement and regular margins, while GBM and SM demonstrated heterogeneous ring enhancement, necrosis, and hemorrhage. The CHAID decision tree achieved an overall accuracy of 80.6%, with class-specific accuracies of 83.3% for GBM, 86.1% for SM, and 91.7% for PCNSL. Conclusions: The proposed decision tree offers a practical tool for diagnostic orientation and treatment planning, particularly in settings without access to advanced imaging techniques.