The Evaluation of Radiomic Models in Distinguishing Pilocytic Astrocytoma From Cystic Oligodendroglioma With Multiparametric MRI.

PURPOSE

To assess whether a machine-learning model based on texture features extracted from multiparametric magnetic resonance imaging could yield an accurate diagnosis in differentiating pilocytic astrocytoma from cystic oligodendrogliomas.

MATERIALS AND METHODS

The preoperative images from multisequences were used for tumor segmentation. Radiomic features were extracted and selected for machine-learning models. Semantic features and selected radiomic features from training data set were built, and the performance of each model was evaluated by receiver operating characteristic curve and accuracy from isolated testing data set.

RESULTS

In terms of different sequences, the best classifier was built by radiomic features extracted from enhanced T1WI-based classifier. The best model in our study turned out to be the gradient boosted trees classifier with an area under curve value of 0.99.

CONCLUSION

Our study showed that gradient boosted trees based on texture features extracted from enhanced T1WI could become an additional tool for improving diagnostic accuracy to differentiate pilocytic astrocytoma from cystic oligodendroglioma.