Rob Mannino demonstrates the image-analysis algorithm-based smartphone app that he developed as a biomedical engineering student at Georgia Tech. The app is capable of detecting anemia just by looking at pictures of a patient's fingernails.Image Credit: Georgia Tech
Every month from the time he was six months old, Rob Mannino has had to go into a clinic to receive a blood transfusion. Mannino has an inherited blood disorder known as beta-thalassemia, which is caused by a mutation in the beta-globin gene.
“My doctors would test my hemoglobin levels more if they could, but it’s a hassle for me to get to the hospital in between transfusions to receive this blood test. Instead, my doctors currently have to just estimate when I’m going to need a transfusion, based on my hemoglobin level trends," Mannino said.
For his biomedical engineering PhD project at Georgia Tech, Mannino set out to develop an image-analysis algorithm capable of detecting anemia just by looking at pictures of a patient's fingernails.
"Every single one of us at some point is at risk for anemia," said principal investigator Wilbur Lam. "Rob has essentially developed a way in which anybody now can answer the question 'am I anemic' and all they need is a smartphone."
Lam is a clinical hematologist-bioengineer at the Aflac Cancer and Blood Disorders Center of Children’s Healthcare of Atlanta, associate professor of pediatrics at Emory University School of Medicine and a faculty member in the Wallace H. Coulter Department of Biomedical Engineering at Georgia Tech in Atlanta, GA.
The researchers published their results Dec. 4, 2018, in Nature Communications.
“All other point-of-care anemia detection tools require external equipment and represent trade-offs between invasiveness, cost, and accuracy,” Lam said. “This is a standalone app that can look at hemoglobin levels without the need to draw blood.”
The app should be used for screening, not a clinical diagnosis, the researchers caution. Mannino was a graduate research assistant in biomedical engineering who has since graduated.
“This whole project couldn’t have been done by anyone but Rob,” Lam said. “He took pictures of himself before and after transfusions as his hemoglobin levels were changing, which enabled him to constantly refine and tweak his technology on himself in a very efficient manner. So essentially, he was his own perfect initial test subject with each iteration of the app.”
The app could facilitate self-management by patients with chronic anemia, allowing them to monitor their disease and to identify the times when they need to adjust their therapies or receive transfusions, the researchers said. That may reduce side effects or complications of having transfusions too early or too late.
The technology could be used by anyone at any time and could be especially appropriate for pregnant women, women with abnormal menstrual bleeding, or runners/athletes. Its simplicity means it could be useful in developing countries. Clinical diagnostic tools have strict accuracy requirements, but Mannino and Lam said that with additional research, they can eventually achieve the accuracy needed to replace blood-based anemia testing for clinical diagnosis. The current gold standard for anemia diagnosis is known as a complete blood count (CBC) test.
The researchers studied fingernail photos and correlated the color of the fingernail beds with hemoglobin levels measured by CBC in 337 people: some healthy, and others with a variety of anemia diagnoses. The algorithm for converting fingernail color to blood hemoglobin level was developed with 237 of these subjects and then tested on 100.
The researchers were able to show that a single smartphone image, without personalized calibration, can measure hemoglobin level with an accuracy of 2.4 grams/deciliter with a sensitivity of up to 97%. Personalized calibration, tested on four patients over the course of several weeks, can improve the accuracy to 0.92 grams/deciliter, a degree of accuracy on par with point-of-care blood-based hemoglobin tests. Normal values are 13.5-17.5 grams/deciliter for males and 12.0-15.5 grams/deciliter for females.
In the app, the use of fingernail beds, which do not contain melanin, means the test can be valid for people with a variety of skin tones. The accuracy is consistent for dark or light skin tones, Mannino said. The app uses image metadata to correct for background brightness and can be adapted to phones from multiple manufacturers.
Mannino and Lam are working with a variety of doctors at Children’s Healthcare of Atlanta and Emory – geriatric, internal medicine, neonatologists, transfusion medicine, global health – to obtain additional data and better calibrate their system.
“This is just a snapshot of the accuracy right now,” Lam said. “The algorithm gets smarter with every patient enrolled.”
The smartphone anemia app is projected to be available commercially for public download as early as Spring 2019. A patent application has been filed for the anemia app, and both Lam and Mannino have a financial interest in the success of the product.