Implementation of an artificial neuronal network to predict shunt necessity in carotid surgery

In carotid surgery, it could be useful to know which patient will tolerate carotid cross-clamping in order to minimize the risks of perioperative strokes. In this clinical study, an artificial neuronal network (ANN) was applied and compared with conventional statistical methods to assess the value of various parameters to predict shunt necessity. Eight hundred and fifty patients undergoing carotid endarterectomy for a high-grade internal carotid artery stenosis under local anesthesia were analyzed regarding shunt necessity using a standard feed-forward, backpropagation ANN (NeuroSolutions (R); NeuroDimensions, Gainesville, FL) with three layers (one input layer, one hidden layer, one output layer). Among the input neurons, preoperative clinical (n = 9) and intraoperative hemodynamic (n = 3) parameters were examined separately. The accuracy of prediction was compared to the results of a regression analysis using the same variables. In 173 patients (20%) a shunt was used because hemispheric deficits or unconsciousness occurred during cross-clamping. With the ANN, not needing a shunt was predicted by preoperative and intraoperative parameters with an accuracy of 96% and 91%, respectively, where the regression analysis showed an accuracy of 98% and 96%, respectively. Those patients who needed a shunt were identified by preoperative parameters in 9% and by intraoperative parameters in 56% when the ANN was used. Regression analysis predicted shunt use correctly in 10% using preoperative parameters and 41% using intraoperative parameters. Intraoperative hemodynamic parameters are more suitable than preoperative parameters to indicate shunt necessity where the application of an ANN provides slightly better results compared to regression analysis. However, the overall accuracy is too low to renounce perioperative neuromonitoring methods like local anesthesia.