Evaluation metric of smile classification by peri-oral tissue segmentation for the automation of digital smile design

Abstract

Objectives: This study aimed to develop and validate evaluation metric for an automated smile classification model termed the "smile index." This innovative model uses computational methods to numerically classify and analyze conventional smile types.

Methods: The datasets used in this study consisted of 300 images to verify, 150 images to validate, and 9 images to test the evaluation metric. Images were annotated using Labelme. Computational techniques were used to calculate smile index values for the study datasets, and the resulting values were evaluated in three stages.

Results: The smile index successfully classified smile types using cutoff values of 0.0285 and 0.193. High accuracy (0.933) was achieved, along with an F1 score greater than 0.09. The smile index successfully reclassified smiles into six types (low, low-to-medium, medium, medium-to-high, high, and extremely high smiles), thereby providing a clear distinction among different smile characteristics.

Conclusion: The smile index is a novel dimensionless parameter for classifying smile types. The index acts as a robust evaluation tool for artificial intelligence models that automatically classify smile types, thereby providing a scientific basis for largely subjective aesthetic elements.

Clinical significance: The computational approach employed by the smile index enables quantitative numerical classification of smile types. This fosters the application of computerized methods in quantifying and analyzing real smile characteristics observed in clinical practice, paving the way for a more objective evidence-based approach to aesthetic dentistry.