Login / Signup

Prescription eyeglasses as a forensic physical evidence: Prediction of age based on refractive error measures using machine learning algorithm.

Maram Atef MoustafaSara Attia GhitaniMarwa Abdelfattah KholiefBassam Ahmed El-Sayed AbulnoorMennattAllah Hassan Attia
Published in: Journal of forensic sciences (2024)
Refractive errors (RE) are commonly reported visual impairment problems worldwide. Previous clinical studies demonstrated age-related changes in human eyes. We hypothesized that the binocular RE metrics including sphere and cylinder power, axis orientation, and interpupillary distance (IPD) can be used for forensic age estimation of an unknown individual. RE data of both eyes were collected from the clinical optometric exams and prescription glasses of 2027 Egyptian individuals aged between 2 to 93 years. The differences between age groups as well as sides, and sexual dimorphism were explored. Two modeling methods were compared: multiple and stepwise linear regression (LR) versus machine learning Regression Forest (RFM). Data were apportioned into training and test datasets with a ratio of 80/20. The results showed significant differences among the age groups in each eye for all variables. Stepwise LR improved the results over models based on the one-sided lens due to selection of IPD in addition to the left and right axis, and left sphere as independent variables. For the RFM, the left axis and IPD were the most important features. RFM outperformed LR in terms of accuracy and root mean squared error (RMSE). The estimated age within ±10 years showed 81.4% accuracy rate and RMSE = 8.9 years versus 38.5% accuracy rate and RMSE = 17.99 years using RFM and stepwise LR, respectively, in the test set. The current study upholds the significance of the age-related changes of refractive error in formulating alternative forensic age estimation models when standard methods are unavailable.
Keyphrases
  • machine learning
  • mental health
  • cataract surgery
  • endothelial cells
  • big data
  • emergency department
  • deep learning
  • climate change
  • quality improvement
  • adverse drug
  • virtual reality
  • pluripotent stem cells
  • drug induced