LDA-Guided Search Engine for the Nonsubjective Analysis of Infrared Microscopic Maps

Abstract

Acquisition of large data sets from human tissues by infrared (IR) microscopy is now routine. However, processing such large data sets, which may contain more than 10 000 spectra, provides an enormous challenge. Overcoming this challenge and developing nonsubjective methods for the analysis of IR microscopic results remain the major hurdle to developing clinically useful applications. A three-step pattern recognition strategy based upon linear discriminant analysis has been developed for use as a search engine for tissue characterization. The three-step strategy includes a genetic algorithm-guided data reduction step, a classification step based upon linear discriminant analysis, and a final step in which the discriminant coefficients are converted into a visually appealing, nonsubjective representation of the distribution of each class throughout the tissue section. The application of this search engine in the characterization of tumor-bearing skin is demonstrated.