Application of classification models to pharyngeal high-resolution manometry.

Area of Science:

• Gastroenterology
• Biomedical Engineering
• Computational Biology

Background:

• Pharyngeal high-resolution manometry (HRM) generates complex spatiotemporal plots.
• Interpreting these plots for swallowing disorders can be time-consuming and subjective.
• Objective, automated analysis methods are needed to enhance clinical utility.

Purpose of the Study:

• To evaluate three pattern recognition methods for analyzing pharyngeal HRM data.
• To assess the accuracy of these models in identifying disordered swallowing patterns.
• To determine the most effective parameters for classification.

Main Methods:

• Employed classification models: artificial neural networks (ANNs) including multilayer perceptron (MLP) and learning vector quantization (LVQ), and support vector machines (SVM).
• Trained models using HRM data from 12 healthy controls and 13 subjects with swallowing disorders, who ingested 5-ml water boluses.
• Validated model performance on independent datasets, analyzing parameter contributions.

Main Results:

• MLP achieved the highest classification accuracy (96.44%), followed by SVM (91.03%) and LVQ (85.39%).
• Parameters related to the upper esophageal sphincter were identified as most significant for accurate classification.
• All tested models demonstrated high performance in differentiating normal and disordered swallowing patterns.

Conclusions:

• Classification models, particularly MLP, offer a highly accurate and efficient method for analyzing pharyngeal HRM data.
• These computational approaches can significantly reduce the time needed for data interpretation in clinical settings.
• Automated pattern recognition facilitates the broader application of HRM in diagnosing and managing swallowing disorders.