Reliability of Comprehensive Facial Soft Tissue Landmark Detection and Analysis Using Frontal View Photographs

Facial soft tissue analysis plays a vital role in orthodontics, orthognathic surgeries, and forensic anthropology. This study aimed to determine a standardized midline to assess the reliability of symmetry analysis in frontal view photographs. Additionally, it evaluated the consistency of soft tissue landmark detection and analysis between different observers and over time.

A power analysis determined the need for a total of 50 standardized frontal-view photographs. These photographs were obtained from Iranian patients aged 9-40 who were undergoing orthodontic treatment. Thirty-seven standardized landmarks were annotated across the face, facilitating 73 distinct facial analyses. Five midlines were compared to identify the most reliable one for symmetry assessments. A Python-based automated framework was created to calculate the analyses based on landmarks' x and y coordinates. Inter-observer and intra-observer reliability were assessed using Intraclass Correlation Coefficient (ICC) and Mean Absolute Deviation (MAD) metrics.

Among the 5 midlines for symmetry analysis, the perpendicular of the interpupillary line passing through the glabella was found to be the most reliable. Overall, the analyses demonstrated relatively good reliability. However, certain landmark points, specifically Gonion (Go), Malar Eminence (ME), Zygion (Zy'), Frontotemporale (Ft'L), and Trichion (Tr'), exhibited lower reliability. Consequently, the reliability of specific analyses, including Bizygomatic width/Facial width, Nasal bridge width/ Endocanthions' distance, and horizontal symmetry of Go', N', and Zy' was reduced.

This study presents a comprehensive collection of soft tissue landmarks, facial analyses, and their corresponding inter- and intra-observer reliability. Furthermore, the most reliable facial midline for symmetry analysis was determined. This study demonstrated that certain indices defined in three dimensions have lower reliability when drawn in 2-dimensional images, affecting the reliability of derived analyses.