Optimized Computational Instrumentation in Biomedicine: A Review

In this correspondence, we review optimized computational instrumentation in biomedicine. Kernel classifiers and novelty detectors are very promising new techniques for signal recognition. However, improvements of their generalization ability can still be achieved by incorporating prior knowledge of recognition task at hand. In this correspondence, we present a review of optimization techniques for wavelet decompositions which can be used as optimal feature extractors for kernel based classification as well as novelty detection. For this, we apply a lattice parameterization of paraunitary filter banks for deriving an optimal representation of the data in features spaces induced by reproducing kernels in the sense of statistical learning theory. We apply our hybrid approach to signal recognition problems in audiology. In particular, we present a powerful approach to the objective detection of the central auditory processing disorder and a novelty detection scheme for identifying otoacoustic emissions.