Skip to content Skip to main navigation Report an accessibility issue


Machine Reading Comprehension Could Improve Medical Visits, Experience

Greg Peterson

Greg Peterson

In an era where it seems increasingly hard to find a consensus on anything, one thing that unifies everyone is a disdain for medical visits that seemingly last hours.

Now, thanks to the joint effort of researchers from the Min H. Kao Department of Electrical Engineering and Computer Science, Oak Ridge National Laboratory, and the Department of Veteran Affairs, those times could be drastically reduced.

The team, which includes EECS Department Head Greg Peterson, joint Oak Ridge National Laboratory (ORNL) faculty member Edmon Begoli, and doctoral student Maria Mahbub, has explored a way of using computer-assisted “reading comprehension” to help machines quickly scan medical documents and extract critical data. This research effort was led by Maria Mahbub whose area of doctoral research is biomedical machine reading comprehension and is working under contract with ORNL as a Student Research Collaborator.

Edmon Begoli

Edmon Begoli

“Biomedical machine reading comprehension, when implemented in computer-assisted clinical decision support systems will allow clinicians to make decisions in a fraction of the time it would take to manually sort through information,” said Mahbub. “This is accomplished by developing applications that can not only ‘read’ the information, but also understand it well enough to provide the clinician with a set answer or possible answers to the problem they are trying to solve.”

The team also includes ORNL’s Alina Peluso as well as Susana Martins and Suzanne Tamang of the Department of Veterans Affairs Office of Mental Health and Suicide Prevention. Tamang is also a member of the Stanford University School of Medicine.

Funding for the project was provided by Department of Veterans Affairs Office of Mental Health and Suicide Prevention and acquired by ORNL. The research used resources provided by ORNL’s Knowledge Discovery Infrastructure.

Their work was presented in the Institute of Electrical and Electronics Engineers (IEEE) Xplore publication under the title, “cpgQA: A Benchmark Dataset for Machine Reading Comprehension Tasks on Clinical Practice Guidelines and a Case Study Using Transfer Learning.” It can be read in full here.

At the heart of the matter are what are known as clinical practice guidelines, the “cpg” in the publication’s name. CPGs are the current recommendations in biomedical literature that relate to patient care.

While there are people with the ability to scan pages at a rapid pace, a medical setting means there can be no room for delays or errors.

By developing machine reading comprehension applications, the speed can be ramped up by an order of magnitude while at the same time offering a higher level of accuracy than a speed-reading person might provide.

“Through our work, we have developed, for the first time, a benchmark dataset for machine reading comprehension on clinical practice guidelines when it comes to the biomedical field, as well as evaluating that dataset,” said Mahbub. “Our hope is that this will work as an important initial step toward introducing machine reading into clinical settings, alleviating what is currently a major problem.”

An achievement that could save time and lives.