Classification of psychotropic drugs in a high dimensional space: Some preliminary results on hypothesis stability and power function

In this paper, we propose an approach to classify psychotropic drugs from the events related potential (ERP) signals using the P300 components. The difficulties of the problem reside essentially in the fact that traditional methods do not apply when observations are in a high dimensional space, which is a common case in biomedical engineering. Our objective is to propose new hypothesis tests that give p-values reflecting the reality of the efficacy criterion of drugs. Our test is based on a pattern recognition approach. We first study the stability of different training algorithms. We then exhibit a relationship between stability and power functions of the corresponding tests. We finally apply our method to test the efficacy of Lorazepam versus placebo to modify generators of brain activity.