Potential breakthrough in infant biometrics heads for trial in Kenya after ID4Africa appeal
Author: huifan Time: 2018-06-29
Infants born to mothers with HIV have between a 15 and 45 percent chance of contracting the virus if they are not treated, but treatment of mothers and babies with antiretroviral drugs at intervals can reduce the transmission rate to below 5 percent, according to World Health Organization statistics. Many infants are lost to treatment programs, however, as their caregivers fail to bring them back for the necessary repeated vaccinations, leading to numerous preventable cases of HIV, causing a range of negative health outcomes, including early death. Identifying those infants who require treatment is a possible first step toward ensuring they receive it, but identifying infants without the proactive help of caregivers is a social problem for which the solution has long eluded the world, including the biometrics industry.
Paul Macharia earned his Masters’degree from the Africa Nazarene University for his work researching infant biometrics in an effort to solve this problem, and improve the rates of infant HIV treatment in Kenya.
“Right now the whole world is talking about elimination of mother to child transmission of HIV,” Macharia told Biometric Update in an interview. “But for that to work, the infants have to come to the clinic, at least for the first 12 months, up to 18 months. Sadly, many mothers or caregivers don’t bring their kids.”
Research unveiled by Anil Jain of Michigan State University at ID4Africa 2017 showed that fingerprints collected at six months of age can be used for identity verification, and Macharia hypothesizes that fingerprints are adequately formed and stable enough for biometric use from birth, but challenges unique to infants have historically prevented the collection of good enough quality data. He says that characteristics such as the moisture content of infant’s skin seem to increase the difficulty posed by the fine ridges of their fingerprints.
To build his case and work toward a resolution, Macharia ran a trial of 80 children in four different age cohorts, all below one year old. The trial showed promise for improving the quality of the biometric data captured, as measured by NFIQ scores, to the point where it would be good enough to be used for identification of the babies, Macharia says. He was able to use biometric data collected from children at only six weeks old to identify them over time, but he reports that some of the worst quality biometric data was collected from older infants, as squirmy nine-month old children are often too excited to fully cooperate with healthcare practitioners.
“It was and still is to an extent an academic approach, but it’s solving a social problem that governments need to solve,” Macharia points out. “When infant biometrics work, it won’t only remain in use for HIV. Civil registration wants to register everyone from birth, and immigration wants to be able to use biometrics to manage all people who are travelling across borders.”
Different continent, same problem
Kleeman' s team at UC San Diego had already been working on biometric identity verification for infants with the support of the Gates Foundation. Kleeman was originally asked to help tackle the challenge of infant biometrics several years ago, and convened a team of nearly 30 experts from various fields.
Kleeman says they solved 80 percent of the problem conceptually in a series of meetings, and then spend four months watching infants to begin bridging the gap from conceptual to practical problem-solving. Early prototypes included a ball with several embedded optical sensors, which Kleeman’s team hoped would collect good quality finger and palm prints as it was grasped.
Kelley' s team has developed a device which registers prints from 10 fingers and three areas of the palm with a non-contact optical sensor. It captures monochrome images, illuminated with multiple different colors of light and polarizations, in an area shielded from external light sources by a hood attachment which can be switched based on the subject and print being taken. It takes 100 images per second, and uses machine learning and pattern recognition to identify and enhance the best images for feature extraction. Rather than storing raw templates, the device hashes them, and then uses a fairly standard biometric matching algorithm, Kelley explains. That produces an accuracy rate of nearly 100 percent for infants 2 weeks and older, and around 90 percent with newborns. It also records infant vital signs non-intrusively, increasing its utility for health care practitioners. It is also inexpensive, to the extent that Kelley believes that shipping them to Kenya may cost more than the devices themselves.
The device is undergoing concept validation testing, so far successfully, with a large-scale vaccination program in a Tijuana hospital. It is being used by healthcare practitioners.
That next round will help build Macharia’s scientific case for the feasibility of infant identification with biometrics, as well as provide more data for training the algorithms, Macharia says. They will also demonstrate the device’s effectiveness and cost-efficiency for other applications, including private sector for-profit projects, which Kelley hopes will help subsidize future humanitarian projects.
Macharia has similar ideas, and sees a wide range of potential uses for the device in health care, but also in financial services, telecommunications, and government services. “For all these industries biometrics is a key component,” he says. “Functional biometrics for unique identification solves a lot of social problems in those industries.”
With so much at stake for so many people, from new mothers with HIV to national governments, the world will be rooting for his success.
