Artificial Intelligence Malaria Rapid Diagnostic Test Reading Under Variable Lighting and Camera White Balance: A Mobile Vision Robustness Evaluation of Healthcare Triage Equity

Mobile vision systems that read malaria rapid diagnostic tests (RDTs) are increasingly proposed as a way to standardise interpretation and reduce observer error in low-resource triage. Their performance is usually reported under controlled illumination and a fixed camera pipeline, yet deployed devices encounter fluorescent wards, direct sunlight, kerosene lamps and automatic white balance that shifts the colour of the test line itself. the present analysis treats that gap as a measurement problem rather than an engineering inconvenience. It develops a formal model in which the observed optical density of the control and test lines is a function of the true analyte concentration, the illuminant spectrum, the camera white-balance gains and sensor noise, and it shows that the decision threshold learned under one illuminant is not invariant under another. The analysis derives the bias in estimated analyte concentration induced by white-balance mismatch, establishes conditions under which a classifier calibrated on a reference illuminant retains its sensitivity across a declared lighting envelope, and characterises the equity consequence: because test-line intensity interacts with haemoglobin and haematocrit, the same lighting shift produces larger reading errors in anaemic patients, who are disproportionately children and pregnant women in malaria-endemic settings. The contribution is a robustness specification for RDT-reading vision systems that names the calibration, colour-constancy and reporting requirements needed before such systems can be used in triage, together with a set of testable propositions for their evaluation.