Journal Information
Vol. 15. Issue 3.
Pages 248-250 (July - September 2022)
Share
Share
Download PDF
More article options
Visits
4185
Vol. 15. Issue 3.
Pages 248-250 (July - September 2022)
Scientific letters
Open Access
Estimating the prevalence of heterochromia iridum from high-resolution digital yearbook portraits
Visits
4185
Matthew Dabkowskia,
Corresponding author
matthew.dabkowski@westpoint.edu

Corresponding author.
, John Casea, Ian Klooa, Julie Pickettb
a United States Military Academy, Department of Systems Engineering, Building 752 (Mahan Hall), 4th Floor, West Point, NY 10996, United States
b United States Military Academy, Keller Army Community Hospital, 900 Washington Road, West Point, NY 10996, United States
This item has received

Under a Creative Commons license
Article information
Full Text
Bibliography
Download PDF
Statistics
Full Text
Introduction

Heterochromia iridum (HI), or complete heterochromia, is a condition where an individual's irises are of two different colors.1 It may be congenital, with or without a syndromic relationship, or acquired through injury, medication, or other means.2 Immaterial of its origin, HI is generally thought to be harmless and easily diagnosable. For these reasons, HI receives little attention in the medical literature, especially when it is not associated with a more concerning medical condition or cause. For example, a PubMed search on June 19, 2021 returned just 16 results for HI and complete heterochromia among its 32 million citations, and all 16 articles headlined HI's relationship with a syndrome, abnormality, or disease.3 Not surprisingly, a precise measure of HI's prevalence remains an open question, although it is considered rare.

While it is possible HI is benign and presents no real medical risk, its association with troubling underlying medical conditions “warrants investigation".4 Moreover, it is difficult (if not impossible) to study potential correlations between HI and other more serious diseases without a better understanding of its prevalence. To the best of our knowledge, the most expansive attempt at estimating HI's prevalence was undertaken by Stelzer, who personally examined over 25,000 Viennese people between 1 and 80 years old in the late 1960s.5 Notably, he only considered cases without a presumed pathological cause, and he reported an observed prevalence of 0.06313%.5

Material and methods

To augment the limited body of knowledge on HI's prevalence, we estimate it using a collection of high-resolution digital yearbook portraits of United States Military Academy (USMA) cadets. As a secondary analysis of publicly available, subsequently deidentified data, USMA’s Institutional Review Board determined this research met 32 CFR § 219.104′s requirements for exempt status.6

Demographically, our sample's subjects have homes of record that abide the geospatial distribution of US congressional seats, and they are generally between 20 and 26 years old. They are also healthy. Specifically, the US Department of Defense Medical Examination Review Board medically screens cadets prior to entering USMA. This extensive process includes an eye examination. After admission, cadets receive periodic health assessments throughout their tenure at USMA, and a significant injury or change in health can trigger medical separation. Finally, eye color tends to stabilize by 6 years of age with minimal changes up to early adulthood.7 Taken together, HI cases in our sample are likely congenital, unassociated with other diseases, and perpetual.

Procedurally, the study followed four steps: (1) photo preparation, where the eyes were identified and extracted from each image using a modern machine learning approach; (2) preliminary identification, where non-experts flagged possible cases; (3) final determination, where an optometrist confirmed or denied the possible cases; and (4) statistical analysis, where prevalence was estimated.

While Steps (2) through (4) are self-explanatory, Step (1) merits additional discussion. Specifically, our algorithm applies Zhang et al.’s Multitask Cascaded Convolutional Networks model, which takes a photo as an input and outputs a set of coordinates representing five facial landmarks, including the eyes.8 For each photo's eye coordinates, we subsequently crop the image twice – once for the left eye and once for the right eye, where each cropped image is a square centered on the respective eye's coordinates. Based on the standard composition of the yearbook portraits, the square's side length was easily tuned to a constant that ensured the iris was contained in each cropped image, and these ocular photos were subsequently arranged in subject pairs. Step (1) was automated using the Python programming language, allowing us to focus our manual efforts on HI identification.

Results

Ultimately, 11,111 photos were processed, and 7 cases of HI were confirmed for an observed prevalence of 0.06300%. Based on its small value, we quantified the uncertainty of our estimate with a 95% Agresti-Coull confidence interval (CI)9 built using R's binom package,10 which yielded (0.02761%, 0.13291%). Our point estimate is nearly identical to Stelzer's result, and our CI contains it. Based on the latter, we fail to reject the null hypothesis of their equality at a significance level of 0.05. That said, it is possible we missed subtle cases in Step (2); therefore, our estimate is potentially conservative.

Discussion

Previously, the most ambitious attempt at estimating HI's prevalence was Stelzer's.5 Despite being the largest such study, it is over 50 years old, drawn exclusively from Vienna residents, and includes young children whose eye color was potentially still developing. Our work affirms Stelzer's result using a geographically diverse, certifiably healthy sample of young adults with more stable iris coloration. This finding is important to the future study of HI, including possible associations between HI and more serious underlying conditions. Additionally, this study creates the opportunity to extend the machine learning methodology to classify HI from images. Such work could be expanded beyond HI to detect other visible eye disorders.

Disclaimer

The views expressed herein are those of the authors and do not reflect the position of the United States Military Academy, the Department of the Army, or the Department of Defense.

Declaration of Competing Interest

None.

References
[1]
McKusick V.A., Kelly J. Heterochromia iridis. Online Mendelian inheritance in man, https://www.omim.org/entry/142500; 2001 [accessed 24 May 2021].
[2]
D. Turbert.
Heterochromia.
[3]
National center for biotechnology information. PubMed, https://pubmed.ncbi.nlm.nih.gov; 2021 [accessed 19 June 2021].
[4]
J. Saniasiaya.
Heterochromia iridis: more than beautiful eyes.
Postgrad Med J, 96 (2020), pp. 721
[5]
O. Stelzer.
Heterochromia iridis: variations in form, age changes, sex dimorphism.
Anthropol Anz, 37 (1979), pp. 107-116
[6]
National Archives and Records Administration. 32 CFR § 219.104 - exempt research. Code of Federal Regulations,https://ecfr.federalregister.gov/current/title-32/subtitle-A/chapter-I/subchapter-M/part-219; 2018 [accessed 19 Jun 2021].
[7]
L.Z. Bito, A. Matheny, K.J. Cruickshanks, D.M. Nondahl, O.B. Carino.
Eye color changes past early childhood: the Louisville Twin Study.
Arch Ophthalmol, 115 (1997), pp. 659-663
[8]
K. Zhang, Z. Zhang, Z. Li, Y. Qiao.
Joint face detection and alignment using multitask cascaded convolutional networks.
IEEE Signal Process Lett, 23 (2016), pp. 1499-1503
[9]
L.D. Brown, T.T. Cai, A. DasGupta.
Interval estimation for a binomial proportion.
Stat Sci, 16 (2001), pp. 101-133
[10]
Sundar D. binom: binomial confidence intervals for several parameterizations. R package version 1.1-1 [software] 2014. https://CRAN.R-project.org/package=binom.
Journal of Optometry
Article options
Tools

Are you a health professional able to prescribe or dispense drugs?