Development and validation of an AI-driven tool to evaluate chewing function: a proof of concept

Area of Science:

• Biomedical Engineering
• Artificial Intelligence in Healthcare
• Oral Health Research

Background:

• Masticatory function is crucial for oral and general health.
• Objective assessment of chewing function remains a clinical challenge.
• Existing methods for evaluating masticatory function are often not clinically feasible.

Purpose of the Study:

• To develop and validate an automated tool for evaluating masticatory function.
• To establish a proof of concept for AI-driven masticatory performance analysis.
• To create a feasible solution for objective clinical assessment of chewing ability.

Main Methods:

• Utilized YOLOv8, a deep neural network, for food fragment detection and segmentation.
• Assessed model performance using bounding box recall, segmentation metrics, and confusion matrix.
• Employed Bland-Altman diagrams to evaluate segmentation performance in physical units.

Main Results:

• YOLOv8 model achieved over 90% recall and sensitivity in detecting food fragments.
• Successfully classified 301 out of 316 ground truth fragments with high accuracy.
• Bland-Altman analysis indicated general agreement, with a slight overestimation in fragment size measurement.

Conclusions:

• Artificial intelligence offers a reliable approach for automated masticatory performance analysis.
• The developed AI tool serves as a valuable asset for future clinical assessments of masticatory function.
• This study demonstrates the potential of AI in advancing objective evaluation of chewing ability.