Facial landmarking of 3D facial images. V1

• Research type

  Research Study

• Full title

  Landmarking of 3D facial images

• IRAS ID

  296003

• Contact name

  Bodore Khalifa Al-baker

• Contact email

  bkkalbaker@dundee.ac.uk

• Sponsor organisation

  The University of Dundee

• Duration of Study in the UK

  1 years, 6 months, 17 days

• Research summary

  Accurate placement of facial landmarks on 3D facial images is important to ensure precise facial measurements and analysis for use in diagnosis and management of craniofacial cases. The manual digitization of the landmarks is time consuming for clinicians. Existing automated landmarking on the facial soft tissue lacks the accuracy and the simplicity for the routine clinical use. To our best knowledge, no previous study has evaluated the automated landmarking performances in subjects with craniofacial deformities.
  In this study, we explore the possibility of developing an automated landmarking algorithm, based on retrospective 3D facial images of various craniofacial deformities, to significantly improve facial landmark localization accuracy in those population. This is a research and development study aiming to develop an acceptable algorithm to accurately locate facial landmarks to facilitate the future analysis of the facial morphology. There is no intention to carry out testing and evaluation of the developed algorithm as a clinical tool, as it is beyond the scope of this study.
  An important task related to the development of automatic landmarking system is to establish a ground truth (gold standard) of facial landmark positions. Manually labelled landmarks are usually considered as the gold standard. They are provided to these algorithms as training datasets and the algorithms learn to automatically generate the same landmarks for unseen test images. Gold standard data are used to estimate the accuracy in facial landmark localisation performance of the developed algorithm. Therefore, in this study we will manually digitise landmarks on existing high quality 3D facial images to establish a gold standard. Statistical facial models and deep learning networks will be trained and validated on the manually digitised data set. Sophisticated algorithms based on the recent advances in computer vision will be combined to deliver the automated facial landmarking in craniofacial anomalies 3D facial images.

• REC name

  East of Scotland Research Ethics Service REC 1

• REC reference

  21/ES/0042

• Date of REC Opinion

  28 Apr 2021

• REC opinion

  Favourable Opinion