An automatic system for the analysis and classification of esophageal motility records

Abstract

Signal processing techniques as well as feature extraction and pattern classification criteria were utilized to develop a system that automatically classifies esophageal motility records into normal and different abnormal cases. The system consists of four parts: processing the recorded signal to remove noise interference, automatic isolation of the different parts of the esophagus, extracting a set of features that quantifies the records, and a classifier to discriminate the different cases. Classification was accomplished using a two-level classifier. A multilayer feedforward neural network trained using the backpropagation algorithm was utilized. Classification of the tubular part and the lower esophageal sphincter was performed separately. The results have shown that 97.4% and 100% correct classification were obtained for the tubular body and the lower sphincter, respectively. It is concluded that the adopted techniques are highly relevant to esophageal data and that the approach followed is feasible and can become a powerful tool for automatic esophageal diagnosis.