Nonparametric discriminant analysis applied to medical diagnosis

The authors present an application of their method for discriminant analysis (Proc. 13th Internat. Conf. on Pattern Recognition, vol. 2, p. 60-4, 1996) to the diagnosis of the neurological diseases hemorrhages and infarction due to ischemia. The method searches for the discriminant directions which maximize the Patrick-Fisher (PF) distance between the projected class-conditional densities. It is a nonparametric method, in the sense that the densities are estimated from the data. Since the PF distance is a highly nonlinear function, the authors use a recursive optimization procedure for searching the directions corresponding to several large local maxima of the PF distance. The application to the medical dataset indicates the potential of the authors´ method for finding a sequence of directions with significant class separation