Application Of Brain Age Prediction Model Based On Convolution Neural Networks In Patients With Cerebral Palsy

Objective:To construct a brain age prediction model using two-dimensional convolutional neural technique without segmenting the brain white matter and brain gray matter of cranial magnetic resonance,to predict the brain age of children with cerebral palsy using the brain age prediction model,to analyze the trend and degree of difference in brain development of children with cerebral palsy,and to further explore other clinical values of brain age difference.Methods:1)A brain age prediction model was constructed using T1-weighted sequences of cranial MRI images of 3474 healthy individuals screened from international public databases.Randomly selected 2442 MRI data were used to train the model,and the remaining 1032 healthy human MRI data were used as the test set.The training set MRI data and the test set MRI data were independent of each other.And Mean absolute error(MAE),correlation coefficient r,coefficient of determination R~2,and root mean squared error(RMSE)were used to evaluate the accuracy of the model.2)455children with cerebral palsy who met the inclusion and exclusion criteria were selected,and 442 healthy peers with similar physiological age to the case group were matched from international published data.The pre-processed MRI data of children with cerebral palsy and healthy peers were included in the brain age prediction model for brain age measurement,and their brain age gap estimation(Brain AGE)were calculated.The Brain AGE between children with cerebral palsy and healthy peers was analyzed,as well as the effects of disease staging,gender,etiology,and cervical perivascular sympathectomy(CPVS)on brain development of children with cerebral palsy.3)Children with cerebral palsy with language dysfunction and CPVS in the second part were followed up,and those with language function recovery grade less than one were defined as the poor prognosis group.Finally,according to the outcome,univariate analyses were performed to screen candidate independent variables by Brain AGE,serum albumin,severity of motor impairment,pairs of combined mental retardation,serum alkaline phosphatase,side of motor impairment,gestational week and birth weight,and candidate specific variables were subjected to the least absolute shrinkage and selection operator(LASSO)regression and binary regression,respectively.The prediction model of poor language function after CPVS in children with cerebral palsy with language dysfunction was constructed by regression with LASSO and binary multivariate logistic regression analysis,and the model was cross-validated with ten folds.To evaluate the performance of the prognostic prediction model.Results:1)The MAE of the training set was 1.85 years old,RMSE=2.39,r=0.99,R~2=0.98;and the MAE of test set was 3.98 years old,RMSE=5.60,r=0.95,R~2=0.90.2)The data of Brain AGE in cerebral palsy group conformed to normal distribution(P=0.076),and the Brain AGE in cerebral palsy group was higher than that in healthy control group(P<0.0001).The actual brain age of children with cerebral palsy was higher than their physiological age(P<0.0001).The Brain AGE of 5-year-old to adult cerebral palsy patients was negatively correlated with their physiological age(P<0.01).The Brain AGE of spastic cerebral palsy was higher than that of mixed cerebral palsy(P<0.01).The Brain AGE of patients with spastic bilateral paralysis was higher than that of patients with spastic unilateral paralysis(P<0.05).In patients with mixed cerebral palsy,the Brain AGE of male patients was higher than that of female patients(P<0.05).Excluding patients with unknown etiology,in female mixed cerebral palsy,the Brain AGE of patients with non-asphyxia was higher than that of patients with asphyxia(P=0.12),the difference was not statistically significant.The Brain AGE one year after CPVS was lower than that before operation(P=0.58),and the difference was not statistically significant.3)In univariate analysis,there were significant differences in the severity of dyskinesia(P<0.001),Brain AGE(P<0.01),serum albumin(P<0.05)and intelligence level(P<0.05)between the two groups.After LASSO regression,when the minimum mean square errorλis 0.011 and the minimum distance standard errorλis 0.052,the variables of the corresponding models are:serum albumin+serum alkaline phosphatase+mental retardation+Brain AGE+motor impairment severity.The severity of preoperative dyskinesia,Brain AGE,asphyxia,serum alkaline phosphatase,serum albumin and mental retardation were analyzed by binary multivariate Logistic regression analysis(forward,stepwise method):Brain AGE(OR=5.357,P<0.001),serum albumin(OR=3.924,P=0.005),motor disturbance severity(OR=5.276,P=0.003).It is an independent predictor of poor language function after CPVS in children with cerebral palsy with language dysfunction.Model 1 and model 2 were constructed according to the results of LASSO regression and bivariate multivariate Logistic regression analysis.Model 1:serum albumin+serum alkaline phosphatase+mental retardation+Brain AGE+severity of dyskinesia;model 2:Brain AGE+serum albumin+severity of dyskinesia.The average AUC and F-measure of model 1 and model 2 are 0.73and 0.82 respectively,and the average AUC and F-measure of model 2 are 0.74 and 0.79respectively.Conclusion:1)the brain age prediction model of 2D convolution neural network between 5 and 86 years old without dividing white matter and gray matter is feasible.2)the brain age of children with cerebral palsy is higher than their physiological age,and the Brain AGE of children with cerebral palsy is higher than that of their healthy peers.Children with cerebral palsy have partial brain atrophy or whole brain atrophy and brain aging after brain damage.The Brain AGE of children with cerebral palsy over 5years old was negatively correlated with their physiological age;the ability of female children with cerebral palsy to return to normal development track after brain damage was better than that of male children with cerebral palsy;the Brain AGE of children with spastic cerebral palsy was higher than that of mixed cerebral palsy,and the Brain AGE of children with bilateral spastic cerebral palsy was higher than that of unilateral spastic cerebral palsy.3)Brain AGE,serum albumin and severity of dyskinesia are independent predictors of language dysfunction prognosis in children with cerebral palsy after CPVS.Brain AGE can reflect the recovery potential of language function in children with cerebral palsy after clinical intervention.Combined with Brain AGE,the prognostic model of language dysfunction in children with cerebral palsy with language dysfunction after CPVS has the potential of clinical application.