Related Experiment Videos
MRI-Based Machine Learning to Differentiate True Progression from Pseudoprogression in Glioblastoma: An Updated
Xin Zhang1, Xiaopeng Guo2, Yu Wang2
1School of Biomedical Engineering & Imaging Sciences, King's College London, London, United Kingdom (X.Z., K.W., T.C.B.).
Background:
Differentiating true progression (TP) from pseudoprogression (PsP) after chemoradiation for glioblastoma remains clinically difficult. We evaluated externally validated diagnostic performance of magnetic resonance imaging (MRI)-based machine learning models and appraised study quality and transparency.
Methods:
PubMed, Embase, MEDLINE, and Cochrane Library were searched through October 1, 2025 (PROSPERO registration CRD420251144894). Eligible adult glioblastoma studies used MRI-based machine learning to distinguish TP from PsP against histopathology and/or clinicoradiologic follow-up. The primary meta-analysis included external test cohorts with extractable 2 × 2 data; internal and combined internal and external test analyses were secondary. Diagnostic accuracy was synthesized with bivariate random-effects models.
Results:
Thirty-one studies including 2,344 patients undergoing imaging assessment were included. Seven external test cohorts, including 219 patient observations, entered the primary meta-analysis. Pooled sensitivity was 0.78 (95% confidence interval [CI], 0.63-0.88), specificity was 0.75 (95% CI, 0.63-0.84), and summary area under the receiver operating characteristic curve (AUC) was 0.81 (bootstrap 95% CI, 0.69-0.91). Internal test cohorts showed higher apparent performance than external cohorts (AUC, 0.91 vs 0.81), but the difference was not statistically significant (difference, -0.09; 95% CI, -0.22 to 0.02; P = 0.12). Risk of bias was frequently high or unclear, and reporting transparency was heterogeneous.
Conclusions:
MRI-based machine learning showed moderate-to-good external validation performance for TP-versus-PsP discrimination, but evidence remains limited by small external test cohorts, bias, and incomplete reporting. Larger transparent multicenter validations are needed before routine clinical implementation.