Back

Display color accuracy and the D15 color arrangement test: A modeling study

Arpadffy-Lovas, T.; Toth-Molnar, E.

2024-11-15 ophthalmology
10.1101/2024.11.15.24314633 medRxiv
Show abstract

Color vision testing is often considered a resource-intensive examination, requiring an anomaloscope, or at least color plates and specialized lighting equipment. Multiple novel, digital color vision testing algorithms have also been described, but were only verified using calibrated displays. In this work, we assessed if color vision testing could be performed with any computer display using two arrangement type color vision tests, the D15 and desaturated D15 panels. Established mathematical models of human color vision deficiencies (CVDs) and publicly available colorimetric data of real world computer screens were used to model how the panel colors would be displayed on computer screens and then perceived and arranged by subjects with varying levels of CVD. A total of 627,431 arrangements were evaluated. We found that all of the modeled displays, including ones with a low color accuracy of deltaE=6, were characterized by specificity above 86%, and the sensitivity of the desaturated D15 panel remained above 85% as well, with negative predictive values above 90% to detect any form of CVD compared to using the unmodifed colors. These findings suggest that any digital screen could be utilized for color vision screening, even without calibration, enabling reliable high-throughput screening utilizing equipment that is already present at virtually all healthcare providers. Since 8% of the population is affected by some form of CVD, simplifeid screening would unlock notable benefits for all stakeholders. Furthermore, the described algorithm could also be used to optimize other digital color vision tests for general use.

Matching journals

The top 4 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.