GRNN usage for medical diagnosis and design circuit blocks

Generalized Regression Neural Network (GRNN) is a memory-based network that provides estimates of continuous variables and converges to the underlying regression surface. Owing to one pass learning algorithm, GRNN is faster than iterative methods. The estimate converges to the conditional mean regression surfaces as more and more examples are stored in the network structure. Important parameters, which determine the performance of an artificial neural network are the ability of generalization, whether or not it can be used for several purposes, the complexity of the network and the suitability for electronic implemantations. GRNN structure is widely used in system modelling and system identification applications. Obtaining successful results with data sets that has large number of samples, shows that this structure can effectively be used in medical diagnosis. Layered and highly parallel structure with one-pass learning leads to fast training and a simpler hardware design. In this work, GRNN structure is experimented on different medical problems and the performance is examined. For this purpose, using TÜBİTAK- YİTAL 1.5u. CMOS process parameters, current mode analogue integrated circuit sub blocks of a GRNN based classifier that can possibly be developed designed. These sub blocks are subtracter circuit, squarer/divider circuit, exponential circuit and multiplier/divider circuit respectively. Keywords: General regression neural network, Medical diagnosis, Subtracter circuit, Squarer/divider circuit, Exponential circuit, Multiplier/divider circuit. x