Improving MRI-based differentiation of brain metastasis, glioblastoma, and primary CNS lymphoma using deep learning.

objectiveTo develop and validate a deep learning (DL) model based on conventional MRI for preoperative differentiation of brain metastasis (BM), glioblastoma (GBM), and primary central nervous system lymphoma (PCNSL), and to evaluate its diagnostic utility as a decision-support tool for radiologists.

methodsThis multicenter retrospective study included 1,298 patients with histopathologically confirmed BM (n = 426), GBM (n = 456), or PCNSL (n = 416). A 2.5D ResNet50-based DL model was developed using axial T2-weighted and contrast-enhanced T1-weighted MRI. A clinical model (CM) incorporating demographic and semantic imaging features was constructed using multivariable logistic regression. Model performance was evaluated in independent internal and external test sets using the area under the receiver operating characteristic curve (AUC) and accuracy with 95% confidence intervals (CIs). Pairwise AUC comparisons were performed using the DeLong test. A reader study involving three radiologists with different experience levels compared diagnostic accuracy before and after DL assistance.

resultsThe DL model achieved AUCs of 0.893 (95% CI: 0.773–1.000) internally and 0.863 (95% CI: 0.862–0.864) externally. One-versus-rest AUCs were significantly higher than those of the CM for GBM and PCNSL in both datasets (all p < 0.001), whereas no significant difference was observed for BM externally (p = 0.97). With DL assistance, mean radiologist accuracy increased from 69.8% to 81.4% internally and from 73.8% to 87.8% externally (all p < 0.05).

conclusionThe proposed DL model based on conventional MRI enables accurate differentiation of BM, GBM, and PCNSL and provides additive diagnostic benefit when integrated into radiological interpretation.