Interpretation and knowledge discovery from a MLP network that performs low back pain classification

Abstract

Artificial neural networks are being increasingly used as medical decision support tools but are currently undermined by their inability to explain or justify their output classifications. Using a new method published by Vaughn (1996), the paper discovers the key inputs used by a multilayer perceptron (MLP) network that diagnoses low back pain (LBP). The knowledge learned by the MLP network from an input training case is expressed as a data relationship from which a valid rule can be directly induced, obviating the need for a combinatorial search based approach. The validation of the data relationships and rules, with the assistance of domain experts, provides a method for validating the MLP network. The aim of the paper is to discover the key inputs that a LBP MLP network uses to classify selected training case examples using the knowledge discovery method and to present the top ranked key inputs that the LBP MLP uses to classify all training cases for each diagnostic class. It is shown how the validation of the top ranked key inputs by the domain experts can lead to the validation of the LBP network.