LDA-guided search engine for the nonsubjective analysis of infrared microscopic maps

Acquisition of large data sets from human tissues by infrared (IR) microsco py is now routine. However, processing such large data sets, which may cont ain more than 10 000 spectra, provides an enormous challenge. Overcoming th is challenge and developing nonsubjective methods for the analysis of IR mi croscopic results remain the major hurdle to developing clinically useful a pplications. A three-step pattern recognition strategy based upon linear di scriminant analysis has been developed for use as a search engine for tissu e characterization. The three-step strategy includes a genetic algorithm-gu ided data reduction step, a classification step based upon linear discrimin ant analysis, and a final step in which the discriminant coefficients are c onverted into a visually appealing, nonsubjective representation of the dis tribution of each class throughout the tissue section. The application of t his search engine in the characterization of tumor-bearing skin is demonstr ated.