Prediction of progesterone receptor expression in high-grade meningioma by using radiomics based on enhanced T1WI

To predict progesterone receptor (PR) expression of high-grade meningioma using radiomics based on enhanced T1-weighted imaging (WI).

There were 157 cases of high-grade meningioma in the study. Seventy-eight cases had negative expression and 79 cases had positive expression. Spearman's rank correlation coefficient and least absolute shrinkage and selection operator (LASSO) regression were used to select the valuable features. The models were developed by naive Bayes (NB), random forest (RF), and support vector machine (SVM). Receiver operating characteristic (ROC) and decision curve analysis (DCA) analysis were used to assess the models.

Nine features were selected as the valuable features using Spearman's analysis and LASSO regression. The RF and NB models achieved the same area under the ROC curve (AUC) of 0.75, which was higher than that of SVM (0.74). There was no significant difference among the AUCs of the three models (p>0.05). There was a larger net benefit in the RF model than the SVM and NB models across all threshold probabilities in the DCA analysis.

The RF model had good performance in predicting PR expression of high-grade meningioma. PR expression evaluation for high-grade meningioma would be helpful in clinical practice.