Purpose: Radiomic features of gliomas are often used to predict genetic markers from radiological studies. Radiomic features were extracted from the contralateral brain to test if tumor texture is driving the predictive power of machine learning models. Ideally, these contralateral models would be a negative control for tumor radiomics models, since many studies use contralateral normal appearing white matter for normalization. This study used those features to attempt to predict IDH mutation status, MGMT promoter methylation, TERT promoter mutation, and ATRX mutation status with random forests.

Methods: Radiomic features were extracted from the tumor region, a mirrored contralateral region, a spherical region within the tumor, a spherical region on the contralateral, and a spherical sample of the ipsilateral side. These features were used independently to predict IDH, MGMT, TERT, and ATRX using random forests.

Main Findings: Contralateral features alone were as predictive of IDH mutation status as tumor features and had predictive power for several genetic markers.

Conclusion: Normalization with contralateral brain should be done carefully, and further investigation of potential radiological changes to the contralateral is warranted.