Comparison of Artificial Intelligence and Clinicians With Different Experience Levels in Assessing Gingival Phenotype

Gingival phenotype, representing the thickness and morphological characteristics of the gingival soft tissues, plays a critical role in periodontal health, treatment planning, and the long-term stability of clinical outcomes. A thin phenotype is associated with increased risk of gingival recession, papilla loss, and inflammatory complications, while a thick phenotype offers better soft tissue stability but may mask inflammation. Accurate and reproducible assessment of gingival phenotype is therefore essential in clinical dentistry.

The periodontal probe transparency method is considered the gold standard for phenotype assessment due to its simplicity and non-invasiveness. In this method, a periodontal probe is inserted into the sulcus from the buccal aspect, and if the probe is visible through the gingival tissue, the phenotype is classified as thin; if not visible, it is classified as thick. However, the method is susceptible to variability depending on examiner experience, lighting conditions, and subjective interpretation.

Visual assessment, which relies solely on the inspection of gingival and tooth morphology in photographs without a probe, offers a non-contact alternative but is similarly subject to examiner-related variability. These limitations highlight the need for standardized and objective approaches to phenotype determination.

Artificial intelligence (AI), particularly deep learning-based image analysis, has shown promising results in dental diagnostics, enabling automated classification of clinical images with high accuracy and reproducibility. In periodontal research, AI has been applied for lesion detection and radiographic interpretation, but its application in gingival phenotype assessment-especially using the probe transparency method and visual assessment-remains unexplored.

This observational study aims to compare the diagnostic performance of a deep learning-based AI model with human examiners of different experience levels (periodontology residents vs. dental students) in assessing gingival phenotype from standardized intraoral photographs using both the periodontal probe transparency method and visual assessment. The reference standard will be the classification provided by an experienced periodontologist using the probe transparency method in a clinical setting.

The study will evaluate and compare accuracy, sensitivity, specificity, and inter-/intra-examiner agreement across examiner groups and the AI model. The findings are expected to provide insights into the potential of AI as a standardizing tool, reducing inter-examiner variability and supporting clinical decision-making, particularly for less experienced clinicians. Additionally, the study may inform the integration of AI-assisted diagnostic tools in dental education and practice, improving training efficiency and clinical outcomes.