Gummy Smile and Artificial Intelligence

Detailed Description

Among the causes of gummy smile (GS) are:

Excessive vertical development of the anterior maxilla Hyperactivity of the perioral muscles, leading to excessive lip elevation during smiling Short clinical crown length, which increases susceptibility to GS Altered passive eruption, gingival inflammation, and gingival hyperplasia contribute to a reduced clinical crown length. Cases where the vertical dimension of the teeth is shorter than the horizontal dimension suggest that passive eruption is a contributing factor to GS.

The extent of GS is not homogenous across all teeth involved in the smile and often results from multiple etiological factors. When gingival display is 3 mm, it is attributed solely to excessive gingival tissue. If an additional 2.2 mm of gingival display is present (3 mm + 2.2 mm), vertical maxillary excess (VME) is considered a contributing factor. When an additional 2.8 mm of gingival display is observed (3 mm + 2.2 mm + 2.8 mm), a hypermobile (HM) lip is implicated.

For instance:

A 3 mm GS over the maxillary left central incisor may be due to HM lip alone

A ≥8 mm GS over the maxillary left second premolar may result from a combination of:

HM lip (2.8 mm) Excessive gingival tissue (3 mm) VME (2.2 mm) This highlights the importance of measuring GS at each tooth level for accurate diagnosis and personalized treatment planning.

Assessing asymmetric GS requires individual measurements for each tooth, which can increase error likelihood and patient chair time. Given the significance of scientific research in identifying personalized intervention strategies for esthetic smile design, artificial intelligence (AI) could help transform the subjective nature of aesthetic perception into an individualized, scientific framework by adhering to specific reference parameters. The use of precise and valid measurements in GS evaluation can facilitate:

Diagnosis Etiological assessment Treatment planning By establishing a comprehensive dataset, the contributions of various etiologies can be analyzed by comparing measured variables with ideal reference values. A meticulous analysis of etiopathogenetic factors and the severity of the condition can be achieved through such an approach. Previous research indicates a need for more studies validating the accuracy and sensitivity of GS diagnosis and its etiological assessment.

AI-Based Analysis and Dataset Formation To date, no study has been found in the literature evaluating gingival display using AI. Unlike studies that use test datasets or publicly available open-access datasets for accuracy scoring, the data obtained in this research are exclusively derived from a training dataset.

To optimize the validation strategy of the AI model, selected clinical metrics were measured separately for each tooth involved in GS. Labeling was performed using a licensed web-based annotation platform that allows for multiple annotation options.

This study will include patients with a high smile line, selected from those who apply to the Department of Periodontology, Faculty of Dentistry, Gazi University. The labeling of patient photographs will be performed by two researchers:

A periodontologist A senior expert in artificial intelligence (AI) \& software development Photographs of patients with gummy smile (GS) will be used to create training, validation, and test datasets.

Methodology for Dataset Formation

The simple random sampling method was used, with:

Confidence level: 0.95 Prevalence: 0.1 Margin of error: 0.05

The formation of these datasets will follow the steps outlined below:

A literature review was conducted to identify standardization methods for gingival visibility measurements. However, no standardized ruler system was found that allows for simultaneous calibration and measurement of gingival display across multiple teeth. Consequently, the need to develop a custom-designed ruler system specifically for this research emerged.

AI Model Training and Data Processing

Gingival display will be determined using a hybrid model. The training process will include:

Pixel Accuracy (PA) Mean Intersection over Union (mIoU) metrics

The dataset will be divided into:

Training set Validation set Testing set

Image preprocessing will include:

Resizing Normalization Transformation into tensors