Disclosures: Jay and another study author report authorship of a patent for a method for noninvasive determination of blood components.
July 29, 2021
3 min read
Save

Algorithm enables detection of anemia from smartphone photos of the inner eyelid

Disclosures: Jay and another study author report authorship of a patent for a method for noninvasive determination of blood components.
You've successfully added to your alerts. You will receive an email when new content is published.

Click Here to Manage Email Alerts

We were unable to process your request. Please try again later. If you continue to have this issue please contact customerservice@slackinc.com.

Anemia is a global public health problem that carries significant risk for mortality and morbidity, particularly among older adults, children and individuals with chronic conditions.

Diagnosis generally involves a complete blood count test. This requires specific lab equipment and trained personnel, including phlebotomists and technicians. Perhaps for this reason, anemia disproportionately affects individuals who live in rural environments, where access to health care is limited.

Infogrpahic showing accuracy of algorithm

In response to the need for affordable, accessible and noninvasive point-of-care testing, researchers have developed an algorithm for anemia detection using an everyday technology: the smartphone camera.

The algorithm, evaluated in a study published in PLoS One, yielded an accuracy rate of 72.6% for detecting anemia using a photo of a patient’s lower eyelid.

“We’re simply trying to recapitulate what clinicians already do,” researcher Gregory D. Jay, MD, PhD, professor of emergency medicine and engineering at Brown University and attending physician at Rhode Island Hospital, told Healio. “Clinicians look at the conjunctivae and they assess pallor, which is associated with anemia. We started trying to replicate that process using a digital camera when digital cameras first became available. We’ve been working in this area for some time.”

A self-supporting application

The researchers chose the palpebral conjunctiva in developing their algorithm because it is easily accessible for photographing, it presents no competing colors between blood vessels and the conjunctival surface, it has a very short distance between the surface and blood vessels, and blood flow to this area is not significantly affected by temperature or other environmental factors.

“The long-term endeavor is to make it a self-supporting application that will live in an iPhone,” he said. “We want to develop an algorithm that will take raw images captured by an iPhone and work with the colors of the conjunctiva that are captured. We could then look at the colors and compare them with a lookup table, so we can correlate that spectral data to the actual known hemoglobin.”

Potential for screening

Jay and colleagues conducted the prospective convenience sample study among ED patients at an academic teaching hospital.

For phase 1 — the algorithm derivation phase — researchers collected smartphone images of the palpebral conjunctiva of 142 patients with a broad range of hemoglobin (HBc) levels. They used image-based parameters in stepwise regression analyses to create a prediction model of estimated HBc.

Phase 2 involved construction of a validation model using data of 202 new patients seen at the ED. Researchers tested the final model, developed based on all 344 participants, for accuracy in anemia and transfusion thresholds.

Results of phase 1 showed a significant association between estimated HBc and laboratory-predicted hemoglobin (HBI) slope = 1.07 (95% CI, 0.98-1.15). HBc showed accuracy of 82.9% (95% CI, 79.3-86.4), sensitivity of 90.7% (95% CI, 87-94.4) and specificity of 73.3% (95% CI, 67.1-79.5) for predicting anemia.

In phase 2, the model showed reduced accuracy (72.6%; 95% CI, 71.4-73.8) sensitivity (72.8%; 95% CI, 71-74.6) and specificity (72.5%; 95% CI, 70.8-74.1). Accuracy for low transfusion thresholds (< 7 g/dL) was 94.4% (95% CI, 93.7-95), whereas accuracy for high transfusion thresholds (< 9 g/dL) was 86% (95% CI, 85-86.9).

These results suggest the possibility of a smartphone app that could be used to screen for anemia remotely, according to Jay. He discussed other methods being used to assess pallor.

“There is a competing method out there that looks at fingernails and has been described in the literature,” Jay said. “The problem with looking at fingernails for pallor is that it isn’t a metric of central circulation, so you’re going to be subject to the vagaries of local perfusion.”

None of these experimental methods seems poised to replace blood testing for anemia, Jay said.

“I think we’re still going to have blood drawing in our future,” he said. “This study was all about looking at the accuracy of this algorithm as a potential screening test. This is one of the first major steps toward that end, but there is a lot of ongoing development.”

Use of a potential smartphone app could facilitate remote detection of anemia, thus making diagnosis easier for individuals living in remote areas, Jay added.

“Ideally, this is something that could augment telemedicine,” he said.

References:

For more information:

Gregory D. Jay, MD, can be reached at Brown University 1 Prospect St., Providence, RI 02812; email: gregory_jay_md@brown.edu.