Machine Learning and Retinal Pigment Score: Ethnicity vs. Biology
Author Information
Author(s): Rajesh Anand E., Abraham Olvera-Barrios, Alasdair N. Warwick, Yue Wu, Kelsey V. Stuart, Mahantesh I. Biradar, Chuin Ying Ung, Anthony P. Khawaja, Robert Luben, Paul J. Foster, Charles R. Cleland, William U. Makupa, Alastair K. Denniston, Matthew J. Burton, Andrews Bastawrous, Pearse A. Keane, Mark A. Chia, Angus W. Turner, Cecilia S. Lee, Adnan Tufail, Aaron Y. Lee, Catherine Egan
Primary Institution: University College London Institute of Ophthalmology
Hypothesis
Can a machine learning-derived retinal pigment score provide a more accurate measure of pigmentation than self-reported ethnicity?
Conclusion
The study found that the retinal pigment score (RPS) is a more nuanced measure of pigmentation than ethnicity, showing significant genetic associations.
Supporting Evidence
- The retinal pigment score (RPS) was validated using data from the UK Biobank and EPIC-Norfolk Study.
- RPS showed strong associations with known genetic loci related to pigmentation.
- RPS can help improve the fairness of AI algorithms in ophthalmology.
Takeaway
Scientists created a new way to measure how dark the retina is, which is better than just asking people about their ethnicity. This helps make sure that computer programs used in eye care work well for everyone.
Methodology
The study used a machine learning algorithm to derive a retinal pigment score from color fundus photographs and validated it against demographic and genetic data from large epidemiological studies.
Potential Biases
The reliance on self-reported ethnicity may introduce bias in understanding the biological variability of pigmentation.
Limitations
The study's participants were predominantly self-reported white and European, limiting the generalizability of the findings.
Participant Demographics
The median age of participants was 56 years, with 55% being female and 92% self-identifying as white.
Statistical Information
P-Value
p<0.001
Confidence Interval
95% CI: 0.834, 0.963
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website