Neural Networks For Medicine: Two Cases

Abstract

This paper describes two applications of Neural Network for medicine. The first case attempts to utilize a self-training back propagation net, which is supervised by a zero-crossing edge operator, for edge detection on ophthalmoscopic image. The experimental results show that the network performs as closely as its training operator but with considerable saving in computation if the whole image is processed by the zero-crossing operator. The second case describes a decision support system for stroke diagnosis. This system attempts to emulate the reasoning of human stroke experts using the relationships between anatomical damage and the patient's signs and symptoms. The test data for this study is derived from the Michael Reese Hospital (MRH) Stroke Database which contains information about 566 cases of stroke and transient ischemic attack (TIA).