Assessment Of Brain Age In Children With Rolandic Epilepsy Based On Machine Learning

Objective:We used support vector regression（SVR）to develop brain age prediction models for children with Rolandic epilepsy in order to explore their brain maturity.Methods:（1）T₂WI,T₂-FLAIR and 3D-T₁WI were performed in children with Rolandic epilepsy and healthy controls recruited from April 2014 to September 2020,and intelligence was assessed in children with epilepsy using the Wechsler Intelligence Scale for Children 2nd edition-Chinese Revised.（2）Based on 3D-T₁WI,Free Surfer5.3.0 was used to extract the gray matter（including cortical gray matter and deep gray matter nuclei）thickness,surface area and volume in children both in the case and control groups using the Destrieux Atlas.（3）Brain age prediction models in the control group were developed using the method of SVR based on independent cortical thickness,surface area,volume and the stacked anatomy measures in the control group,which selected by the least absolute shrinkage and selection operator（LASSO）regression.Pearson’s correlation coefficient（r）,coefficient of determination（R²）,mean absolute error（MAE）,and root mean square error（RMSE）between chronological age and predicted brain age were calculated for the control group to evaluate the accuracy of these models.（4）The SVR brain age prediction models of the control group were used to predict the brain age of children with Rolandic epilepsy,and the brain-predicted age difference between the predicted brain age and the chronological age and difference analysisalso be conducted.Statistical analysis was performed using IBM SPSS 18.0 version.All the quoted results were two-tailed values,and P<0.05 was considered as statistically significant.Result:（1）Fifty children with Rolandic epilepsy（27 males and 23 females,aged10.1±2.1 years）and 50 healthy control children（30 males and 20 females,aged 10.3±2.1 years）were included in this study,and there was no statistical difference（P>0.05）in gender,age and years of education between the case and control groups.（2）There were significant correlations between the predicted brain age and the chronological age in healthy controls for the models:cortical thickness（r=0.91,R²=0.82,MAE=1.00,RMSE=7.87）,surface area（r=0.63,R²=0.40,MAE=1.61,RMSE=13.81）,volume（r=0.67,R²=0.45,MAE=1.82,RMSE=14.66）,and stacked anatomy（r=0.93,R²=0.87,MAE=0.73,RMSE=5.95）.（3）The results of brain age prediction of children with Rolandic epilepsy by the four SVR models showed that the average brain-predicted age difference was:cortical thickness:-0.1,surface area:-0.3,volume:-0.3,and stacked anatomy:0.3.In addition,there were no statistical difference between the four predicted brain age and the chronological age（P>0.05）.Conclusion:Brain maturity in children with Rolandic epilepsy can be assessed using SVR brain age prediction models based on morphological measures of brain gray matter.