A large scale memory (LAMSTAR) neural network for medical diagnosis

Discusses applications of the LAMSTAR network to a medical diagnostic case; specifically, to a urologic medical diagnosis. The LAMSTAR network is a self trained network based on SOM (Self-Organizing-Map) modules. It employs arrays of link-weight vectors to channel information vertically and horizontally through the network to facilitate fast memory retrieval. For diagnosis, the LAMSTAR network displays the diagnosis with suggestions to perform specific further tests. Also, the network interpolate/extrapolate those subwords (states of car systems), that were not present in the input word. As a medical diagnostic tool, the LAMSTAR network evaluates patients´ conditions and long term forecasting after removal of kidney stones. The LAMSTAR network attempts to predict the treatment´s results (failure/success) by analyzing the correlations among 100 patients (input words), each described by 17 subwords. The paper thus illustrates the scope of applications of the LAMSTAR network