Medical diagnosis by the virtual physician

The purpose of this work is to train a backpropagation neural network to make correct diagnosis of thyroid diseases and to compare its performance with human practitioners of medicine. An 84-14-12 neural network is implemented using 84 signs and symptoms of thyroid diseases as input and the 12 kinds of thyroid illness as output. The training takes place first by varying the number of hidden nodes, then by varying the number of hidden layers, the learning coefficient, the momentum coefficient, the noise inject and the tolerance between output and targets. The training is terminated when the neural network can diagnose all the targeted diseases. A field tested investigation of performance is conducted. The study shows that it is possible to train a "virtual" physician as implemented by a neural network to make correct diagnosis of thyroid diseases based on the signs and symptoms. Such a "virtual" physician outperforms human doctors.