Discrimination between artefacts and contractions in pressure signals from the gastrointestinal tract by pattern recognition method

A pattern recognition algorithm has been developed to discriminate between artefacts and contractions in interdigestive motility recorded by a pressure catheter with four channels from the human duodenum. A learning and a test set, both containing natural and induced artefacts, such as respiration and body movement, are obtained from five volunteers. The event classes were phase I, II and III contractions of the interdigestive motility complex and artefacts from respiration, cough, calibration signals and movements. Length, area, amplitude, inter-event interval, up- and downstroke, and correlation to other pressure channels and to respiration, are applied to classify the events. The sensitivity of the computer scoring increases with the number of applied features. When all the features are applied, the sensitivity of the Bayes' classifier against the visually scored contractions and artefacts is 0.96 with a specificity of 0.69.