Team profile by Dominic Garrity, Adam Wax, Nirmish Shah, Ashley Chompre, Connie Xu, Erin Dollard, Grace Wei, Helen Xu, Isabelle Xiong, Jennifer Teo, Jay Yoon, Kartik Pejavara, Khushmeet Chandi, Kyle Ferguson, Larry Jiang, Olivia Fan, Rayan Malik, Selena Halabi, Sophie Wu, Sunggun Lee and Tingnan Hu

Anemia is a condition in which the number of red blood cells is insufficient to meet the body’s physiological needs and is one of the most common blood conditions in the world, affecting about one third of the world’s population. Pregnant women with anemia experience increased risk of death, while children may experience hindered motor and mental development. 

Regions in low-income countries are highly affected by anemia but lack the resources to screen for it. The most common diagnostic tool used is the hemocytometer, a portable device that calculates hemoglobin from a patient’s blood sample. However, like many other diagnostic tools, it is time- and skill-intensive, expensive and invasive, making it impractical in low-resource settings. 

Shown in the above figure is the team structure for the Bass Connections team over the 2022-2023 academic year. The team leaders include Dominic Garrity, a recent graduate from Harvard University who originally founded the project, Dr. Nirmish Shah in the hematology department of the Duke University Medical Center, and Dr. Adam Wax, a professor in the biomedical engineering department. Garrity was in charge of overseeing the day-to-day project tasks whereas Dr. Shah and Dr. Wax focused on running the clinical study and mentoring the subteams, respectively. 

The students within the project came from a variety of different backgrounds but were separated into two teams for the purposes of this project. The Engineering and Design subteam focused on developing the predictive models as well as a mobile app. The Operations and Outreach subteam focused on fostering domestic and international collaborations. Additionally, this group was in charge of reporting on the team’s progress in the form of manuscript drafting, abstract submission to research conferences and poster presentations.

Using images of nail beds and relevant demographic information from more than 100 patients at the Duke University Medical Center, we developed two predictive models to extract hemoglobin levels from nail bed images: multilinear regression and random forest. Our preliminary results suggest that the random forest model using image processing techniques for the RGC color space is the best predictor of hemoglobin concentration. Aside from this, we fostered a collaboration with Johns Hopkins, University of Pennsylvania and Bangabandhu Sheikh Mujib Medical University in Bangladesh.

We presented posters at the Health Data Science Poster Showcase and the Bass Connections Showcase. We also presented our recent findings as an oral presentation during the 2023 7th International Conference on Medical and Health Informatics Kyoto, Japan. We have also drafted a JMIR protocol paper for submission during the summer of 2023.

A Predictive and Machine Learning Approach to Non-Invasive Anemia Diagnosis

Poster by Dominic Garrity, Adam Wax, Nirmish Shah, Ashley Chompre, Connie Xu, Erin Dollard, Grace Wei, Helen Xu, Isabelle Xiong, Jennifer Teo, Jay Yoon, Kartik Pejavara, Khushmeet Chandi, Kyle Ferguson, Larry Jiang, Olivia Fan, Rayan Malik, Selena Halabi, Sophie Wu, Sunggun Lee and Tingnan Hu