Mendoza and Scalzo Publish New Research on Machine Learning Models

Steve Mendoza and Fabien Scalzo, two Seaver College Data Science professors, recently published their shared research article entitled, “Determining and Validating Population Differences in Magnetic Resonance Angiography Using Sparse Representation,” with the Institute of Electrical and Electronics Engineers (IEEE). This text provides a framework to analyze the inferences made by machine learning models in the realm of diagnostic cerebrovascular radiology. 

“I want this article to visualize the differences between population groups using a more intuitive algorithm,” says Mendoza. “I’ve focused my research on pushing machine learning models to learn the things we need them to learn.”

A machine learning model is a computer program that has been trained to identify specific patterns in a set of data. Based on these patterns, the model will sort all forthcoming inputs, arranging them in an efficient and objective manner. Mendoza and Scalzo specifically evaluated machine learning models in the context of medical imaging, seeking to discover habitual gaps in the program's algorithm. 

Specifically, Mendoza and Scalzo studied data sets emerging from magnetic resonance angiography (MRA), which is an imaging mechanism helpful in diagnosing cardiovascular diseases such as strokes and aneurysms. Cardiovascular complications are among the top five reasons for death or disability within the United States, and there is an apparent gender disparity existing between men and women who suffer such events. For instance, men are more likely to suffer strokes; whereas females are more likely to suffer aneurysms. 

In studying how machine learning models go about sorting data from MRA readings, Mendoza and Scalzo illustrate that these data crunching programs are not infallible. They work off of highly specific generalizations that could vary from person to person. As a result, the Seaver College professors highlight an important point of potential growth for machine learning models.

If you would like to read more about Mendoza’s and Scalzo’s research, visit IEEE’s webpage.