Logistic Discriminant Functions in Electroencephalography

Abstract

Introduction. Neurofeedback treatments in clinical settings are designed to normalize abnormal quantitative EEG (QEEG) features. Typically the patient’s electrophysiological features are compared to a database in order to quantify the patient’s deviance from normative values. A clear diagnosis is crucial in deciding the most appropriate protocol for the case. Combining the initial diagnosis with the database information offers a reliable procedure for deciding the most appropriate neurofeedback protocol. Method. Logistic regression is a powerful model for performing discriminant analysis. Discriminant functions are a tool used to quantify the probability that the patient’s QEEG features are typical of either of two groups. One of the two groups is generally chosen to be normative, while the other is a homogeneous clinical group. While normative databases can detect deviation from normality, they cannot indicate if the QEEG pattern is typically observed in a particular clinical condition. Discriminant functions, however, address these types of issues. Consequently, normative databases and discriminant functions can be considered as complementary tools. With the widespread availability of software which is able to compute the maximum likelihood estimations, logistic regression has recently became a popular tool in biology. However, to date it has received little attention in electrophysiology. With this article we hope to stimulate the application of logistic regression model in this field. The logistic regression discriminant model with a single quantitative predictor is illustrated. Results. The construction of a discriminant function is presented in a step-by-step fashion, stressing methodological, practical and theoretical concerns that should be addressed in order to achieve valid and useful results. A working example illustrates, in a step-by-step fashion, how to implement the discriminant process. Discussion. The approach taken in this article aims to highlight the intuitive appeal and simplicity of this technique. It is shown that the discriminant process can easily be automated with little effort on the part of the clinician.