Student duo using machine learning to improve diabetes treatment in India

// Industrial & Systems Engineering, Biomedical Engineering

Photo of patient in India

A diabetic patient gets a foot examination in India. Submitted photo.

For several thousand dollars, diabetic patients can purchase a mat capable of thermally imaging their feet to preemptively detect ulcers and then transmitting data for remote monitoring.

The technology is impressive. But the cost means it’s also out of reach for most patients and clinics in low-resource settings across the globe, where diabetes is most prevalent.

Photo of Jan Wodnicki
Jan Wodnicki

“Those kinds of resources are just not available to many people in India,” says Jan Wodnicki, who’s one of two University of Wisconsin-Madison engineering students working on a lower cost, portable alternative, with India as the focal point. “They’re kind of left out of this ‘big medicine.’”

Wodnicki, a junior industrial engineering major from Brookfield, Wisconsin, and Thor Larson, a senior biomedical engineering major from La Crosse, Wisconsin, are using a Wisconsin Idea Fellowship to support the project. It’s a blend of medical device prototyping and advanced data analysis.

“We see it as kind of a rapid triage solution, where instead of having everyone assessed by a physician, you would take a picture of their feet and they would get some kind of risk score. If they were deemed low risk, they would be sent back home and if not, they would proceed further with clinical examination,” says Wodnicki. “The motivation is that there is such a high prevalence of diabetes in India and overwhelming demand at hospitals.”

The idea emerged from the Department of Biomedical Engineering’s undergraduate design curriculum via alumna Kayla Huemer (BSBME ’18), who worked on a device prototype as a Fulbright Scholar in India but wanted to incorporate machine learning for data analysis.

Wodnicki and Larson worked more on the mechanical design of the device during their semester together in the biomedical engineering design program. Wodnicki subsequently switched his major, but the two have continued the effort and turned their attention to the data science side of the project.

They’ve created an app to streamline the data collection process when taking images, allowing users to quickly analyze the data. And, with guidance from Industrial and Systems Engineering Assistant Professor Justin Boutilier, Wodnicki and Larson have used data from Huemer’s work in India to develop algorithms that can identify ulcers before they break through the skin. The two took Boutilier’s Machine Learning in Action course during the fall 2020 semester, allowing them to tinker with different methods of data analysis.

Photo of Thor Larson
Thor Larson

They’re currently able to identify, with roughly 89% accuracy, whether a patient has developed an ulcer based on thermal imaging data—and hope to push that rate above 95%. To achieve that—and create an automated process to quickly generate a risk score for each patient—they say they simply need to collect more data in the field, which would allow them to use more advanced machine learning techniques broadly known as deep learning to refine their algorithms.

The COVID-19 pandemic shelved their plans to travel to India in summer 2020, but they’re hopeful they’ll be able to go once conditions improve around the world and travel is less restricted.

“In machine learning, data is everything,” says Larson. “But what’s really important right now and what we’re trying to focus on is what data is actually important. It’s good to know before we go to India what we need to collect, how to collect it in the best way and what format to collect it in.”

Both students say the experience has given them a glimpse into what they hope could be their future careers—somewhere near the intersection of software, data analysis and healthcare.

“I think global health is the perfect place for machine learning,” says Larson. “I hope this project will give me experience in a global health scenario, implementing machine learning in the real world, and then maybe taking that to other projects—either in global health or here in the U.S.—and trying to use machine learning to better healthcare.”

 

Author: Tom Ziemer