Research On Establishment Of Mathematical Models Of Intracranial Hypertension

Objective To explore the methodology on forcasting model of intracranial pressure separately with auto regressive integrated moving average(ARIMA),back propagation neural networks(BPNN) and nonparameter stepwise discriminant analysis (NSDA),and to evaluate the simulation effects.Methods Intracranial pressure(ICP) measured with epidural sensor, mean velocity of middle cerebral artery(VMCA), mean arterial pressure(MAP), partial end-tidal CO2(PETCO2) and heart rate(HR) were recorded continuously for intracranial hypertension(ICH) patients.The ICP time series data of ICH patients were used to establish ARIMA model with SAS software and to fit the ICP variation process of prediction;The data of ICP and 4 independent variables were used to establish BPNN model with Matlab software, and the calibration and predition test were carried out. Finanly ICP and 4 independent variables data were also analyzed with NSDA of SAS software to establish a semi-quantity model, and the discriminant tests were applied for forecast of ICP classification.Results Each of the ICP predictive errors simulated by ARIMA model was within±3mmHg. Coefficient correlation of BPNN between predictive value and true value was 0.99, and mean of absolute error and of relative error were 1.17mmHg, 7.36% respectively. According rate of NSDA model was respectively 94.58%, 92.65% and 92.86% while classifying into ICP normal or slight increase rank, middle increase rank and high increase rank, and the mean of the errors was only 6.02%.Conclusion It's approving that ARIMA model can be used to simulate the variztion of ICP by time if the database is precisely recorded. The model of BPNN is correspond to the non-linear characyeristic of ICH, it has a favorable effect on the prediction of ICP. The model of NSDA is also ideal to achieve high percentage of correct discrimination. Therefore all the three models are hopeful for potential clinical use in the future.