Possibility of predicting Ht values during a plasma exchange therapy using back propagation neural network

There is a sickness that is called Guillain-Barre Syndrome(GBS). The GBS has a disorder of peripheral nerve and the muscle is paralyzed. There is a plasma exchange therapy in one of the effective therapy method for GBS. But this therapy uses a large amount of plasma that was gathered from other people. Therefore, it has the possibility of causing the rejection, rapid decrease of blood pressure or headache using medical machine. There are various factors of decreasing blood pressure and one of them is decreasing of hematocrit(Ht) value. Ht value shows the balance of water within a patient's body. At the present Crit - Line Monitor(CLM) is used for obtaining the Ht values during the plasma exchange therapy in the Tokushima University Hospital. Prediction of the Ht value during the plasma exchange therapy is important to prevent from decreasing of the blood pressure. The purpose of this study is to predict using the neural network what Ht value will change in one, three and five minutes during plasma exchange therapy. The moving average neural network was used as a prediction method. The back propagation algorithm was used for the training algorithm of the neural network(BPNN). In this study, three layers and four layers BPNN were used for prediction and the structure of BPNN was optimized each. The best structures of BPNN in three layers and four layers were decided using the rms error between the prediction Ht value and the measurement Ht value. The results in three and four layers were 0.17, 0.24 in one minute later, 0.30, 0.35 in three minutes later and 0.44, 0.39 in five minutes later each. Possibility of predicting Ht values during a plasma exchange therapy using a BPNN could be suggested from these results.