Evaluation of an automated approach for facial midline detection and asymmetry assessment: A preliminary study

Abstract

Objective: To examine the level of agreement between the conventional method and a machine-learning approach to facial midline determination and asymmetry assessment.

Settings and sample population: The study included a total of 90 samples (53 females; 37 males) with different levels of mandibular asymmetry.

Materials and methods: Two researchers placed predefined soft tissue landmarks individually on selected facial frontal photographs and created 10 reference lines. The midsagittal line was determined as perpendicular to the midpoint of the bipupillary line, and the same two reference lines and facial landmarks were automatically determined by the software using machine-learning algorithms, and researchers created the other 8 reference lines using the facial landmarks that were determined automatically by the software. In the following stage, 2 linear and 10 angular measurements were made by a single researcher on 270 photographs, and the consistency and differences between the measurements were evaluated with a one-sample t test, an intraclass correlation coefficient (ICC) and Bland-Altman Plots.

Results: The level of agreement of measurements between the researchers and the software was low for eight parameters (ICC˂0.70). The one-sample t test revealed that differences between the software and researcher measurements of lip canting and pronasale deviation were not statistically significantly different (P > .05). Aside from the body inclination difference in Group 3 (samples with a mandibular body inclination difference >6°), there was no clinically significant difference (˂3°) between the measurements of the two methods.

Conclusions: Machine-learning algorithms have the potential for clinical use in asymmetry assessment and midline determination and can help clinicians in a manual approach.

Keywords: artificial intelligence; facial asymmetry; machine learning.