Preliminary investigation on predicting postoperative glioma recurrences based on a multiparametric radiomics model.

BACKGROUND

The early prediction of postoperative recurrence and high recurrence area of gliomas is important for individualized clinical treatment. This study aimed to evaluate the performance of a magnetic resonance imaging (MRI)-based multiparametric radiomics model for the early prediction of postoperative recurrences.

METHODS

The data from 60 patients who met the inclusion criteria between 2000 and 2021 were collected in this study. Radiological features were extracted from the T1-weighted imaging (T1WI) and T2WI/fluid-attenuated inversion recovery sequence images. The multiparametric model was composed of two classifiers, the support vector machine and the logistic regression (LR), and it was used for training and prediction. The highest scoring classifiers and sequences were screened out according to the area under the curve (AUC) and accuracy.

RESULTS

For predicting the postoperative recurrences and high recurrence areas of gliomas, the performance of the LR classifier was most stable, and the multiparametric model based on clinical information, basic imaging, and radiomics had the best performance (AUC: 0.99; Accuracy: 0.96).

CONCLUSION

The MRI-based multiparametric radiomics method provided a non-invasive, stable, and relatively accurate method for the early prediction of postoperative recurrences, which has guiding importance for individualized clinical treatment.