Mobile Apps and Machine Learning for Noninvasive Anemia Diagnosis (2022-2023)
Anemia, a disease characterized by a lack of healthy red blood cells that impairs tissue oxygen supply, affects about a fifth of the world’s population and is the second leading cause of disability. Individuals in western and central sub-Saharan Africa and South Asia are disproportionately affected by anemia, with children under five years of age and maternal populations experiencing the highest prevalence. In pregnancy, anemia increases the risk of preterm labor, low birth weight and infant and maternal death. In children, anemia may hinder motor and mental development and elevate the risk of infection and cardiac failure.
Blood hemoglobin is an important indicator used to diagnose anemia. In high-income settings, it is measured using hematology tools that automatically provide accurate and reliable measurements. However, in low-income settings, such tools are not accessible due to their high cost. Instead, hemoglobin is measured using simpler devices, which require invasive capillary blood collection. This usually causes discomfort and may not be suitable for pregnant, elderly and pediatric populations. Thus, despite an overall increased ability to measure hemoglobin, several communities still cannot conveniently and accurately screen for anemia at a low cost.
This project team will to develop a mobile application that quickly and accurately estimates hemoglobin levels from patient nail bed images.
In the first phase, team members will construct a dataset that maps hemoglobin levels to patient nail bed images and other variables such as age and gender. Next, they will train several predictive models, such as random forests and neural networks. After doing so, they will determine which model best estimates hemoglobin concentrations from nail bed images.
In the second phase, the team will deploy the best model on a mobile app and then test the accuracy and robustness of its predictions from novel data, comparing them to results of “ground-truth” blood tests. If the data show interesting trends in certain populations (e.g., if the model seems to only be effective for female populations), follow-up research may be conducted.
In the third phase, team members will combine data collected at Duke with data collected elsewhere to create a generalizable model. If successful, they will compare this model to blood tests for various populations and deploy it on their existing mobile app. If not, they will continue to refine their existing model.
In the fourth phase, they will expand the app’s functionality by reviewing literature, conducting needs assessments and adding features that are feasible to implement and valuable to potential users. These features may include ones that allow users to communicate with health care providers, examine their data over time, report their symptoms and/or learn about anemia.
Mobile app that estimates users’ hemoglobin levels from nail bed images; patent for method of hemoglobin measurement; peer-reviewed publications; engineering competition pitches and presentations
Ideally, this project team will be comprised of 3 graduate students and 12 undergraduate students. Interested students will likely be passionate about medicine and technology and interested in mitigating global health inequities. Useful background experience includes app development, machine learning, clinical research and project management. Experience working in low-income settings will also be valuable.
The team will consist of two subteams: engineering and design; and operations and outreach. The engineering and design subteam will gain experience with app development, computer vision, product design and data science. The operations and outreach subteam will gain experience with clinical study design, stakeholder alignment, project management and regulatory processes (including IRB approval). Although students will focus primarily on the goals of their subteam, they can also contribute to the other subteam.
Student team members will meet twice weekly – once for thirty minutes and once for an hour.
Every student will develop skills relevant to human-centered research design and implementation. If empirical findings are significant, students will have the opportunity to write and publish a scientific manuscript, present at research conferences and submit patent applications. If interested, students will also have the opportunity to assist with data collection in clinical settings.
Sophie Wu will serve as the project manager.
Selected students will have the opportunity to travel to Bangalore, India in Summer 2023.
Summer 2022 – Summer 2023
- Summer 2022 (optional): Develop baseline mobile app; train and evaluate predictive models using data collected at Duke; forge new partnerships and collect data at new sites; adapt protocol and apply for IRB approval at each new site
- Fall 2022: Finish first evaluation of optimal model; deploy app as part of initial study; prepare author methods paper if results are substantial
- Spring 2023: Improve and generalize model; test mobile app at new sites; present new findings
- Summer 2023 (optional): Interview potential users of mobile app; identify necessary improvements; prototype and develop new features; visit partners/users abroad; continue to refine model(s); present new findings; seek additional data
Academic credit available for fall and spring semesters; summer funding available
Image: Hemoglobin, by Edumol Molecular Visualizations, licensed under CC BY-NC 2.0
- Dominic Garrity, Global Alliance for Medical Innovation
- Nirmish Shah, School of Medicine-Medicine: Hematology
- Adam Wax, Pratt School of Engineering-Biomedical Engineering
/undergraduate Team Members
Sophie Wu, Biomedical Engineering (BSE)
/zcommunity Team Members
Global Alliance for Medical Innovation
Jennifer Teo, MD Student, Duke-NUS Medical School
Prashanth Thankachan, St. John's Research Institute, Bangalore, India