Facial morphology analysis in osteogenesis imperfecta types I, III and IV using computer vision. (24th May 2021)

• Record Type: Journal Article
• Authors: Rousseau, Maxime; Vargas, Javier; Rauch, Frank; Marulanda, Juliana; Retrouvey, Jean‐Marc
• Other Names: Yamashiro Takashi guestEditor.; Ko Ching‐Chang guestEditor.
• Is Part Of: Orthodontics and craniofacial research. Volume 24(2021)Supplement 2
• Page Start: 92
• Page End: 99
• Publication Date: 2021-05-24
• Subjects: facial analysis -- machine learning -- osteogenesis imperfecta; Skull -- Growth -- Periodicals; Facial bones -- Growth -- Periodicals; Orthodontics -- Periodicals; Orthodontics, Corrective -- Periodicals; Orthodontic appliances -- Periodicals; 617.51
• DOI: 10.1111/ocr.12491
• Languages: English
• ISSNs: 1601-6335

Abstract

Abstract: Objective: Osteogenesis imperfecta (OI) is an autosomal dominant genetic disease that mainly affects the COL1A1/A2 genes. Individuals affected by OI types I, III, and IV have been reported to demonstrate characteristic facial manifestations of the disease. This study aimed to quantitatively assess OI patients' morphological characteristics. Materials and Methods: This retrospective case‐control study involved 306 individuals (145 male and 161 female). It used automatic facial annotation and statistical shape analysis to compare facial photographs of individuals affected by OI types I, III, and IV with a normocephalic control group. Four facial ratios were used to compare facial proportions. Additionally, we proposed a novel approach to facial analysis using 68 landmarks and statistical shape analysis to compare morphological features. A predictive model (PCALog) was trained to detect whether a subject was affected by OI, based on facial landmarks. Results: Our findings correlate with previous reports of OI type III patients' facial characteristics being the most severely affected among the three types studied. Our novel approach facilitated an interpretation and comparison of morphological changes. Moreover, we successfully trained our PCALog model to automatically detect OI based on landmark features. Conclusion: We found patients' facial manifestations of OI to be more pronounced at the level of the eyes and temples. Our morphological approach facilitates the comparison of various groups and should be considered for future craniofacial analysis studies. Machine learning models can be trained using facial landmarks to detect the presence of conditions that affect facial morphology.