Classification of esophageal motility records using neural networks

This paper suggests an automatic diagnostic system for esophageal motility records using neural networks. Signal processing techniques, feature extraction, and pattern recognition criteria were combined to develop computer programs to be used in identifying, characterizing and classifying of esophageal motility recordings. The architecture of such an automated system includes four cooperating modules: a digital filter to remove the interfered noise, separation of peristaltic waveforms from the tubular region of the esophagus, feature extraction module to detect the main quantitative parameters of each esophageal part, and a multilayer feed-forward neural network trained using the conjugate gradient algorithm was used to classify the peristalsis into different categories. The percentage of correct classification reaches 100%.