Sleep apnea is a chronic respiratory disorder that can lead to stroke, hypertension and heart disease. It’s often underdiagnosed for some people. Shumit Saha, Ph.D., assistant professor, biomedical data science.
Thanks to funding from the National Institutes of Health AIM AHEAD program, he plans to address that by developing an explainable machine learning pipeline for sleep apnea diagnosis.
Sleep apnea is typically diagnosed through polysomnography. Those tests provide extensive sensor data, suitable for applying machine learning models for improved diagnosis.
“Unfortunately,” explains Dr. Saha, “many current machine learning models for sleep apnea don’t account for differences in patient populations, often underrepresenting certain groups.”
The project has two main goals. First, Dr. Saha’s team will identify any shortcomings in current models to ensure fair and accurate predictions for all patients. They will use data from six varied sleep studies to build a solid foundation for this analysis. Next, Dr. Saha will refine the models to address any identified issues. This step will include testing the models to determine which factors influence their decisions, ensuring the final tool is both accurate and explainable.
“Our goal was to not only identify potential issues in the algorithms but also correct them through a thorough and comprehensive set of analyses,” says Dr. Saha. “At project end, we aim to present the first responsible and fair machine learning pipeline for sleep apnea diagnosis.”
The project will also involve community engagement through collaboration with local organizations and providing training opportunities for graduate students in biomedical data science and health informatics through Meharry SACS academic programs.
