Using Machine Learning to Shed Light on Cannabis Effects

machine learning

A photo of Dr. Eleonora Tubaldi Cannabis use is on the rise, but much remains to be understood about how the drug affects health—and particularly how it might impact the human cardiovascular system. To help fill in that research gap, UMD mechanical engineering assistant professor Eleonora Tubaldi is partnering with Jean Jeudy, MD, a diagnostic radiology expert and professor at the University of Maryland School of Medicine (UMSOM), and Timm-Michael Dickfeld, MD, a cardiac electrophysiologist expert and professor at UMSOM, in a bid to use machine learning and advanced computing to obtain hitherto elusive answers.

“Our goal is to uncover the hidden links between cannabis use and cardiovascular health,” Tubaldi said. “We aim to determine whether there are morphological changes, changes in the structure or configuration of the heart chamber, or clinical values that are modified because of cannabis use.”

To obtain their findings, Jeudy, Tubaldi and Dickfeld will be making use of UK Biobank, a massive database that includes more than 40,000 images. Tubaldi will develop algorithms and a machine learning framework that will allow a computer to quickly and accurately compare the cardiovascular systems of cannabis smokers to those of non-users. Alternative methods of cannabis intake, such as edibles, are not part of the study.

A machine can often detect changes that are too subtle to be registered by a human observer, Tubaldi said.

“Our goal is to uncover the hidden links between cannabis use and cardiovascular health. We aim to determine whether there are morphological changes, changes in the structure or configuration of the heart chamber, or clinical values that are modified because of cannabis use.”

Dr. Eleonora Tubaldi, assistant professor of mechanical engineering, University of Maryland

“We have limited vision and can see only some aspects of an image,” she said. “Machine learning can tell us about the texture of the image, or about contrast features that even a clinically trained eye might not be able to detect. It’s like a super-eye.”

It can also provide more consistent findings. “If you put ten radiologists in the same room, looking at the same image, they may very well not agree,” she said. “But with machine learning, instead of a person making a guess, we have a machine with an algorithm that knows precisely what to look for and can make a very precise observation. This introduces a level of objectivity that can really be an asset to understanding.”

For now, the team’s focus is on images obtained through a type of MRI known as T1 mapping, which tracks the relaxation rate of a certain tissue over a set time period. If the approach is successful, the scope may ultimately expand to include other types of imaging.

“There is a strong need to study this topic, and to be able to determine how much cannabis can be tolerated by the body,” Tubaldi said. “Such research can assist medical professionals and also inform policymakers.”

The research being undertaken by Jeudy, Tubaldi and Dickfeld is supported by an MPower Seed Grant, awarded by the University of Maryland Strategic Partnership: MPowering the State. The partnership brings together Maryland’s two most powerful public research institutions—the University of Maryland, Baltimore (UMB) and the University of Maryland, College Park (UMCP)—and leverages the sizable strengths and complementary missions of both institutions.

Published March 31, 2022