Using artificial neural network to predict functional recovery of patients treated by intravenous thrombolysis in acute ischemic stroke

In general, cerebrovascular diseases are composed of approximately 80% ischemic strokes. They are both expensive and time consuming while physicians take care of the acute ischemic stroke (AIS) patients. It is well-known that thrombolysis treatment in AIS patients can reduce disability and increase survival rate, however only one-half of patients have good outcomes. Therefore, we designed a functional recovery prediction model by artificial neural network (ANN) for AIS patients after intravenous thrombolysis to help make better clinical decisions. In this study, we retrospectively collected 157 AIS patients who received intravenous thrombolysis at a medical center in north Taiwan. The outcome defined Modified Rankin Scale ≤2 after three-months follow-up as favorable recovery. 80% data were selected for training this predictive ANN model and 20% data were used for validation. The performance of models is evaluated by Receiver Operating Characteristic (ROC) Curve Analysis. An ANN with 5 inputs and 6 neurons in hidden layer was obtained. The performance of this model was with accuracy 83.87% and the area under ROC curve 0.87. This results showed that this ANN model could achieve a high prediction accuracy for functional recovery evaluation. It is an important issue to predict prognosis of treatment for personalized medicine. Risk and benefit should always be balanced before any treatment is to be applied. The developed prediction models may help physicians to individually discuss and explain the likely recovery probability to patients and their families within short therapeutic time before thrombolysis treatment in the emergency room.