One of the most difficult problems in mHealth is patient identification. Villagers in rural areas of developing countries often lack medical health records. Names used are also often very common, which makes it extremely difficult for health workers to accurately identify patients and link them to their health records. Without accurate health records, it can be difficult to properly diagnosis, monitor, and care for patients. Villagers can be uncomfortable with have their fingerprints scanned or head shots taken because they are worried about identity theft.
What it does
Ears are as unique as fingerprints. Our program allows health workers to take pictures of patients' ears and register them with an unique patient ID. Additionally it allows the health workers to search for patient ID by taking a picture of the follow-up patient's ear.
How I built it
The mobile data collection app was built in Commcare because it is a widely used mHealth platform. It syncs its data to our google drive, where Matlab and R crunch the data to extract features from new ears and compare features to existing ears. Matlab then returns the top 10 patients with similar ears. Built an illumination shield to help ensure each ear picture is taken with uniform illumination, rotation, and scale.
Challenges I ran into
Commcare's free servers are very busy during African daytime. Feature extraction takes a lot of time.
Accomplishments that I'm proud of
Tested algorithm against lot of data both from volunteer ears and from existing databases of ear. Have been able to successfully ID several test ears.
What I learned
Commcare is annoying. Ears come in many different sizes. People in the research community should share code more often.
What's next for IdEar
Refine our algorithm by including gender and estimated age into the search algorithm. ODKclinic app development. Better backend database development.