Neurocomputing applications in post-operative liver transplant monitoring

This paper explores the potential for the application of neurocomputing technology to the domain of post-operative liver transplant monitoring. The investigation compares a neural network model with two classical statistical techniques using biochemical information obtained from a set of liver transplant patients. Each approach combines the results of a number of liver function tests to predict the presence of allograft rejection. Each system is assessed, relative to the clinical gold standard, in terms of its overall accuracy and degree of advance warning offered. Applying nonlinear methods does offer an advantage over the traditional linear techniques. The underlying structure of the data set has also been determined using k-means cluster analysis. This analysis suggests important directions for future investigation including the use of temporal information. Preliminary results of incorporating this temporal information are also presented