Objective: To identify the magnetic resonance imaging (MRI) features associated with epidermal growth factor (EGFR) expression level in lower grade gliomas using radiomic analysis.
Methods: 270 lower grade glioma patients with known EGFR expression status were randomly assigned into training (n=200) and validation (n=70) sets, and were subjected to feature extraction. Using a logistic regression model, a signature of MRI features was identified to be predictive of the EGFR expression level in lower grade gliomas in the training set, and the accuracy of prediction was assessed in the validation set.
Results: A signature of 41 MRI features achieved accuracies of 82.5% (area under the curve [AUC] = 0.90) in the training set and 90.0% (AUC = 0.95) in the validation set. This radiomic signature consisted of 25 first-order statistics or related wavelet features (including range, standard deviation, uniformity, variance), one shape and size-based feature (spherical disproportion), and 15 textural features or related wavelet features (including sum variance, sum entropy, run percentage).
Conclusions: A radiomic signature allowing for the prediction of the EGFR expression level in patients with lower grade glioma was identified, suggesting that using tumour-derived radiological features for predicting genomic information is feasible.
Key points: • EGFR expression status is an important biomarker for gliomas. • EGFR in lower grade gliomas could be predicted using radiogenomic analysis. • A logistic regression model is an efficient approach for analysing radiomic features.
Keywords: EGFR; Lower grade glioma; MRI; Prediction; Radiomics.