The application of pattern recognition to the identification of pathogens by laser-excited fluorimetry

✍ Scribed by J.T. Coburn; R.A. Forbes; B.S. Freiser; L. Becker; F.E. Lytle; D.M. Huber

• Publisher: Elsevier Science
• Year: 1986
• Tongue: English
• Weight: 775 KB
• Volume: 184
• Category: Article
• ISSN: 0003-2670
• DOI: 10.1016/s0003-2670(00)86470-3

✦ Synopsis

The k-nearest neighbor (LNN) approach to pattern recognition was used to evaluate a laser-based method of identifying pathogens based on aminopeptidase profiling. Suitability of the method was tested by evaluating the recognition accuracy for differentiation of four bacterial genera as well as four races within a single species. Even though variations in profile replicates were relatively large, the kNN approach successfully differentiated these pathogens. Recognition accuracies of 100% and 92% were achieved for the differentiation of genera and races, respectively. Feature-selection algorithms were used which allowed the rejection of features which did not add useful information towards identification or improve recognition accuracy. Identification of races was facilitated by constructing a data set comprised of only races of one species, because feature selection and weighting were strongly affected by the easily differentiated genera. Methods used for feature selection and weighting were also evaluated.