June 2023
Volume 64, Issue 8
Open Access
ARVO Annual Meeting Abstract  |   June 2023
Open-Source Machine Learning Tool for Craniofacial Photo Recognition
Author Affiliations & Notes
  • George Nahass
    Plastic Surgery, University of Illinois Chicago, Chicago, Illinois, United States
  • Jeffrey C Peterson
    Opthamology, Illinois Eye and Ear Infirmary, Chicago, Illinois, United States
  • Nikki Khandwala
    Opthamology, Illinois Eye and Ear Infirmary, Chicago, Illinois, United States
  • Kevin Heinze
    Opthamology, Illinois Eye and Ear Infirmary, Chicago, Illinois, United States
  • Akriti Choudhary
    Plastic Surgery, University of Illinois Chicago, Chicago, Illinois, United States
  • Chad A Purnell
    Plastic Surgery, University of Illinois Chicago, Chicago, Illinois, United States
  • Ann Q Tran
    Opthamology, Illinois Eye and Ear Infirmary, Chicago, Illinois, United States
  • Footnotes
    Commercial Relationships   George Nahass None; Jeffrey Peterson None; Nikki Khandwala None; Kevin Heinze None; Akriti Choudhary None; Chad Purnell None; Ann Tran None
  • Footnotes
    Support  NIH Grant P30EY001792 (PI: Dr. Deepak Shukla), Research to Prevent Blindness Unrestricted Departmental Grant
Investigative Ophthalmology & Visual Science June 2023, Vol.64, 1106. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      George Nahass, Jeffrey C Peterson, Nikki Khandwala, Kevin Heinze, Akriti Choudhary, Chad A Purnell, Ann Q Tran; Open-Source Machine Learning Tool for Craniofacial Photo Recognition. Invest. Ophthalmol. Vis. Sci. 2023;64(8):1106.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Purpose : To develop an algorithm to automate the assessment of periorbital and craniofacial measurements using a combination of open source and de-novo developed computational tools.

Methods : Photos from the University of Illinois at Chicago Craniofacial Center database were included in this study. Included in the database were front, side and orthodontal images of patients with craniofacial abnormalities. Software developed in-house, FaceFinder, was used to parse the mixed data set by detecting the presence of faces and two eyes using the haar cascade to identify front facing images. To determine data accuracy, human grading was used to assure only front facing images were saved within the new dataset. Additional cropping to capture the facial profile was performed automatically. The metadata from the photos including, but not limited to, focal length, focal length in 35 mm, camera model, f number and exposure time, was automatically stored in the new image. The metadata of each image was also used to ensure the images were saved in portrait orientation. The time from execution of the script to complete was noted.

Results : A total of 6014 images were analyzed by FaceFinder. All folder and file names had non-alphanumeric characters removed automatically before machine learning detection. Of the initial images, 27.3% (n=1642) were identified as face candidates. Our data cleaning algorithm detected facial images with two eyes present at 87.3 +/- 2.8% accuracy (1421/1642 images) as determined by the average of 3 human graders (2 UIC ophthalmology residents and one medical student). In 100% of successful facial detections, facial cropping and metadata copying was successful, and 99.9% of images were saved in portrait mode (1419/1421 images). The execution time on 6014, 3081, 1694 and 895 images was 7.15, 3.52, 2.13 and 1.26 hours respectively.

Conclusions : This machine learning tool is capable of sorting through a heterogenous dataset of craniofacial photos. This capability allows for automated sorting that can facilitate downstream analysis of curvilinear oculofacial features. Improvement on data preparation will allow for future periorbital measurements with artificial intelligence oculofacial mapping to be notably expedited by reducing data preparation time.

This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.

×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×