Artificial intelligence algorithms for differentiating pseudoprogression from true progression in high-grade gliomas:...

Differentiating pseudoprogression (PsP) from true progression (TP) in high-grade glioma (HGG) patients is still challenging and critical for effective treatment management. This meta-analysis evaluates the diagnostic accuracy of artificial intelligence (AI) algorithms in making this distinction. We aimed to assess the performance of AI algorithms in distinguishing between pseudoprogression and true progression in patients with high-grade glioma. We searched PubMed, Cochrane, and Embase databases for studies reporting on AI algorithms that differentiate pseudoprogression from true progression in high-grade gliomas. The analysis evaluated reported metrics such as accuracy, sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and F1 score. The meta-analysis included 26 articles involving 1,972 patients. In the high-grade glioma group, AI algorithms demonstrated a sensitivity of 88% (95% CI: 77%-100%) and a specificity of 75% (95% CI: 54%-97%). For the glioblastoma (GBM) group, accuracy was 77% (95% CI: 68%-86%), with sensitivity of 77% (95% CI: 67%-86%) and specificity of 63% (95% CI: 43%-82%). Overall, the algorithms achieved an accuracy of 80% (95% CI: 76%-85%), sensitivity of 85% (95% CI: 80%-91%), specificity of 69% (95% CI: 58%-80%), a PPV of 79% (95% CI: 58%-100%), a NPV of 97% (95% CI: 90%-100%), and an F1 score of 74% (95% CI: 67%-81%). AI algorithms show significant promise in accurately distinguishing between pseudoprogression and true progression in high-grade gliomas, suggesting their potential utility in clinical decision-making.