Machine learning cuts diagnosis, treatment time for gut disease

The diagnosis and treatment of a gut disease that can cause permanent damage in children can be sped up by the application of machine learning, according to researchers from the University of Virginia schools of Engineering and Medicine.

The researchers used machine learning algorithms on biopsy images from children suffering from environmental enteric dysfunction, which limits the gut’s ability to absorb nutrients and can stunt growth, impair brain development and even lead to death. In low- and middle-income countries, such as Bangladesh, Zambia and Pakistan, the problem is widespread­­––20% of children under the age of 5 are affected by the disease­­––though a population of children in rural Virginia are also affected. Their work was recently published in JAMA Open Network.

Researchers wanted to speed up the diagnostic and treatment process of the disease, which is preventable. They used a convolutional neural network (CNN) to analyze images and diagnose the gut disease in children. The network was trained on duodenal biopsy images, with the data collected, prepared and analyzed between November 2017 and February 2018.

In total, 102 children participated in the study, with the average age of 31 months. The model had a detection accuracy of 93.4%, and a false negative rate of 2.4%. The method was able to detect either environmental enteric dysfunction or celiac disease, which is a common cause of stunting in the U.S.

The study results underscore how pathologists will be able to leverage machine learning tools and AI to more effectively screen patients based on where neural networks are looking for differences and where the tools are focused for their analysis. The machine learning algorithm can offer insights not seen by human eyes, work as a validation system to a pathologist’s diagnosis and reduce the time between imaging and diagnosis, according to researchers.

“There is so much poverty and such an unfair set of consequences,” Sana Syed, MD, MS, assistant professor at the UVA School of Medicine and first author of the study. “If we can use these cutting-edge technologies and ways of looking at data through data science, we can get answers faster and help these children sooner."