Abstract
This paper presents the analysis of the electromyographic signals from rat stomaches to identify and classify contractions. The results were validated with both visual identification and an ultrasonic system to guarantee the reference. Some parameters were defined and associated to the energy of the signal in frequency domain and grouped in a P vector. The parameters were statistically analyzed and according to the results, an artificial neuronal network was designed to use the P vectors as inputs to classify the electrical signals related to the contraction conditions. A first approach classification was performed with and without contraction classes (CR and NCR), then the same database were subdivided in four classes: with induced contraction (ICR), spontaneous contraction (SCR), without contraction due a post mortem condition (PMR) or under physiological conditions (PNCR). In a two-class classifier, performance was 86%, 93% and 91% of detections for each electrogastromyografic (EGMG) signal from each of three pairs of electrodes considered. Because in the four-class classifier, enough data was not collected for the first pair, then a three-class classifier with 82% of performance was used. For the other two EGMG signals electrode pairs, performance was of 76% and 86% respectively. Based in the results, the analysis of P vectors could be used as a contraction detector in motility studies due to different stimuli in a rat model.