Pattern Recognition Techniques Applied to Biomedical Problems

Abstract

Nausea is a common set of symptoms related to several underlying physiological causes, usually difficult to identify a priori. Detecting nausea before emesis (vomiting) is particularly important for patients who are still unconscious after surgery, because emesis may cause various life-threatening complications. Electrogastrography (EGG) is the cutaneous measurement of the electrical activity of the stomach sensed by electrodes placed on the abdomen of the patient. As the relationship between nausea and gastric dysrhythmias is not yet well understood, the study of electrogastrograms may generate information to relate these processes. Thus, the aim of this study was to evaluate the possibility of detecting the presence of nausea in chemotherapy patients. The method consists of acquiring signals using multichannel electrogastrograms, isolating the gastric motion information applying independent component analysis, and then processing the resulting signal to discriminate between normal function and nausea. Feature extraction, clustering, and selection yielded a classifier that discriminated between both classes. The performance of the classifiers was compared among different experiments which include the number of channels and the period of nausea observation (pre- and post-detection). The best classifier obtained 83.33% of accuracy discriminating 31 control and 29 nausea events, using only 4 EGG channels and 3 features: the dominant power, the dominant frequency, and the relationship between the maximum and the average power of the event.