Asymmetry of Three-Dimensional Craniofacial Images and Its Quantified Perception

Project: National Science and Technology CouncilNational Science and Technology Council Academic Grants

Project Details


Scope and Significance. Facial symmetry serves as an important clinical indicator, specifically in craniofacial orthodontics and the plastic surgery, to assess the attractiveness of appearance. Craniofacial features adopted, and limitations of two-dimensional (2D) imaging, such as shades, occlusion, skin textures, etc., all affect the readings. However, qualitative interpretations may easily lead to misunderstanding and challenged objectivity, if without support of quantitative renditions. Perception and expectation gap, as well as opaque information with asymmetrical predominance, between stakeholders – physicians and patients – is the major factor causing medical disputes. An important national health status report in 2012 showed that 26% of domestic medical disputes were arose against surgical departments, which include craniofacial orthodontics and the plastic surgery, in particular. Perfect facial symmetry is considered a theoretical or impractical existence. Most attractive people exhibit asymmetric facial nature. Intentionally mirroring one’s photograph may result in odd look. Even so, significant facial asymmetry can still cause aesthetic and functional problems. Whether facial asymmetry originates from congenital, developmental, or acquired causes, effective assessment and treatment planning and management are crucial topics between the physicians and their patients and need to be well addressed. Objectives. The specific aims of the proposed project are threefold: (1) Quantification of facial asymmetry that is in agreement with the clinical experience; (2) Justification of the “perception buffer zones” from symmetry to serious asymmetry – an overall asymmetry index, or oAI, will be proposed to transparently articulate the perception between stakeholders; (3) Construction of a facial symmetry classifier (that categorizes the perception into Perceived Normal (PN), Perceived Asymmetrically Normal (PAN), and Perceived Abnormal (PA) that implies the necessity of a surgical improvement) along with regarding confidence index (Ci) of the classification result. The consistency between the Tri-information {oAI, PN|PAN|PA, Ci} and clinical readings can be considered as a key performance factor (KPI) of the proposed project. Study Design and Schematic Procedures. (1) Position twenty three-dimensional (3D) facial landmarks that constitute fourteen features; (2) Morph a single normal non-textured craniofacial image (normal occlusion, no craniofacial deformity, no facial trauma history, having not experienced prior orthodontic or orthognathic surgery) to automatically generate a series of slightly to severely asymmetrical “stimulus faces”; (3) Conduct a facial questionnaire survey, Perform pretest validity analysis, Realize perception convergence principle to capture and quantify the awareness of asymmetry; (4) Perform statistical and reliability analysis, Estimate rough oAI; (5) Build a neural network classifier, Formulate oAI parameters, Compute the final oAI; (6) Compile the tri-information {oAI, PN|PAN|PA, Ci} and Implement clinical validation. Academic Contributions. (1) This study will identify and validate 20 facial-symmetry related landmarks with reference of Farkas’ definition in Anthropometry. These 20 landmarks are grouped into 8 medial and 6 pairs of bilateral points, resulting into 14 features. (2) Non-textural three-dimensional CBCT and 3dMD images are employed. CBCT and 3dMD images are of 3D and of high resolution, and thus alleviate difficulties of accurate readings caused by 2D occlusion. Further, use of non-textural images could avoid skin quality or facial attractiveness being intrinsically taken into account. (3) Last, but not the least, in our pilot study of the proposed project, the novel application of detecting rapid changes in concavity intensity over the facial surfaces has shown great efficacy and outperformed many other algorithms. Its academic contributions can be readily anticipated.

Project IDs

Project ID:PB10507-2960
External Project ID:MOST105-2221-E182-013
Effective start/end date01/08/1631/07/17


Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.