Discriminant analysis for anaesthetic decision-making: an intelligent recognition system for epidural needle insertion.

Abstract

BACKGROUND: Incorrect placement of epidural catheters causes medical complications. We used linear discriminant analysis (LDA) to develop an intelligent recognition system (i-RS) in order to guide epidural placement and reduce physician error. METHODS: We analysed real-time dual-wavelength fibreoptic data recorded from the end of an epidural needle in a live porcine model. Two categories of tissue layers were necessary for correct placement of catheter: epidural space and ligamentum flavum. The data were tested using linear, quadratic and logistic parametric analysis to identify which method could distinguish the two anatomical structures. RESULTS: LDA was the best fit for our model. There was ∼80% sensitivity and specificity for correct anatomical identification. Error rates based on cross-validation were 17.0% for the epidural space and 18.6% for ligamentum flavum. Error rates were greater with the 532 nm compared with 650 nm wavelength. CONCLUSIONS: The sensitivity and specificity of LDA for identifying the correct anatomical structure was similar to a physician who is an expert in epidural placement. Overall performance of an i-RS could be improved by expanding the database for decision-making and adding a category of uncertainty. This would reduce complications caused by incorrect epidural placement.