Radiomic features of magnetic resonance images as novel preoperative predictive factors of bone invasion in meningiomas

Abstract

Purpose

Bone invasion in meningiomas is a prognostic determinant, and a priori knowledge may alter surgical techniques. Here, we aim to predict bone invasion in meningiomas using radiomic signatures based on preoperative, contrast-enhanced T1-weighted (T1C) and T2-weighted (T2) magnetic resonance imaging (MRI).

Methods

In this retrospective study, 490 patients diagnosed with meningiomas, including WHO grade I (448cases), grade II (38cases), and grade III (4cases), were enrolled and 213 out of 490 cases (43.5 %) had bone invasion. The patients were randomly divided into training (n = 343) and test (n = 147) datasets at a 7:3 ratio. For each patient, 1227 radiomic features were extracted from T1C and T2, respectively. Spearman's correlation and least absolute shrinkage and selection operator (LASSO) regression analyses were performed to select the most informative features. Subsequently, a 5-fold cross-validation was used to compare the performance of different classification algorithms, and logistic regression was chosen to predict the risk of bone invasion.

Results

Eight radiomic features were selected from T1C and T2 respectively, and three models were built using radiomic features. The radiomic models derived from T1C alone or a combination of T1C and T2 had the best performance in predicting risk of bone invasion, with areas under the curve in the training dataset of 0.714 [95 % CI, 0.660−0.768] and 0.722 [95 % CI, 0.668−0.776] and in the test datasets of 0.715 [95 % CI, 0.632−0.798] and 0.713 [95 % CI, 0.628−0.798], respectively.

Conclusions

The radiomic model may aid clinicians with preoperative prediction of bone invasion by meningiomas, which can help in predicting prognosis and devising surgical strategies.