Classification of postural profiles among mouth-breathing children by learning vector quantization

Area of Science:

• Musculoskeletal health
• Artificial Intelligence in Medicine
• Child Posturology

Background:

• Mouth breathing is a chronic condition linked to musculoskeletal and postural changes in children.
• Identifying characteristic postural patterns in mouth breathers presents a significant challenge.
• Learning Vector Quantization (LVQ) offers a potential AI solution for analyzing these complex postural changes.

Purpose of the Study:

• To apply LVQ for identifying characteristic postural profiles in mouth-breathing children.
• To enhance the understanding of abnormal posture associated with mouth breathing.
• To develop a classification system for mouth breather postural abnormalities.

Main Methods:

• Utilized postural training and validation datasets from 52 and 32 children, respectively (differentiating mouth and nose breathers).
• Compared LVQ performance against various machine learning models including self-organizing maps, back-propagation, Bayesian networks, and k-nearest-neighbor.
• Assessed classifier accuracy using leave-one-out cross-validation, Area Under the ROC Curve (AUC), and Kappa statistics.

Main Results:

• The LVQ model successfully identified five distinct postural profiles specific to mouth-breathing children.
• LVQ demonstrated high classification accuracy for differentiating mouth and nose breathers, with sensitivity and specificity rates of 0.90-0.95.
• The model achieved robust performance on both training and validation datasets.

Conclusions:

• The five identified postural profiles for mouth breathers were integrated into software.
• This application aids in classifying the severity of abnormal posture in children with chronic mouth breathing.
• LVQ proves effective in characterizing and classifying postural deviations in mouth-breathing populations.