Magnetic resonance imaging (MRI)–derived radiomics and deep learning models for predicting epidermal growth factor receptor status in glioma: a systematic review and meta-analysis

AIM: Epidermal growth factor receptor (EGFR) alterations are associated with prognosis in glioma patients. We systematically reviewed the performance of radiomics and deep learning (DL) models in predicting EGFR alterations and provided pooled estimates for the former. MATERIAL AND METHODS: A comprehensive search was performed across PubMed, Scopus, Embase, and Web of Science. Study quality was assessed using the METhodological RadiomICs Score (METRICS) tool. A bivariate model was used to compute pooled estimates of sensitivity, specificity, positive likelihood ratio (DLR+), negative likelihood ratio (DLR-), diagnostic odds ratio (DOR), and area under the curve (AUC). A subgroup analysis was carried out to explore potential sources of heterogeneity. To detect potential outliers, bivariate boxplot and sensitivity analyses were performed. Deek's funnel plot asymmetry test was utilised to evaluate the publication bias. RESULTS: A total of 12 studies assessing EGFR amplification were included in the systematic review, with six incorporated into the meta-analysis. The pooled results of the radiomics models were as follows: sensitivity: 0.74; specificity: 0.88; DLR+: 6.12; DLR-: 0.29; DOR: 21.07; and AUC: 0.87. A significantly higher pooled specificity was observed among studies conducted in Asia than among those performed in North America. No significant publication bias was detected, and the sensitivity analysis did not identify any outliers. CONCLUSION: Radiomics may have potential in predicting EGFR amplification in glioma patients, improving prognostic assessments and treatment planning. These models may facilitate noninvasive patient stratification for EGFR-targeted therapies, potentially improving clinical decision-making and reducing the need for high-risk biopsies.