Deep Learning for Gummy Smile Segmentation

Overview

A gummy smile (excessive visibility of the gums when smiling) is not merely an aesthetic issue but also an important parameter in terms of periodontal health. Current evaluation methods are subjective and non-standardized, leading to limitations in both clinical accuracy and patient communication. In recent years, AI-based models have begun to be effectively used in dental image analysis and diagnostic processes. This study aims to develop an AI-supported objective and reproducible analysis model capable of evaluating gummy smile from both aesthetic and periodontal perspectives using a unique dataset composed of images obtained through standard clinical protocols and labeled by the same expert. Individuals aged 12 years or older with no maxillary anterior (teeth #13-23) tooth loss will be included in the study. Patients with missing anterior maxillary teeth (teeth #13-23), significant anatomical pathologies, or smile-interfering factors (e.g., facial piercings, orthodontic appliances, facial hair) will be excluded. Standardized frontal photographs will be taken using a single device (iPhone 15) to ensure consistency in resolution, lighting, and color balance. Images will be captured from a fixed distance of 15 cm with participants in an upright position, eyes facing forward, and heads aligned to the Frankfurt Horizontal Plane. To maintain standardization, the smartphone's grid lines will be used to align the horizontal line with the pupils and vertical lines with the nasal alae. Images of high, average, and low smile lines will be labeled by a periodontist using the web-based annotation tool MakeSense. Visible gingival areas will be annotated as polygons bounded superiorly by the lower border of the upper lip and inferiorly by the gingival margin. For participants with high smile lines, gingival display will be measured using ImageJ (National Institutes of Health, Bethesda, MD, USA), with calibration performed via a periodontal probe embedded in each photo. A pixel-to-millimeter conversion factor will be derived and applied to measurements between the upper lip and gingival margin in the anterior maxillary sextant (teeth #13-23). Distances between paired landmarks (points 7-13, 8-14, 9-15, 10-16, 11-17, 12-18) will be measured in millimeters. AI-based segmentation outputs (via MakeSense) will be statistically compared to ImageJ measurements to assess correlation.