StFX faculty-student journal article explores AI technology for diagnosis of chronic kidney disease

computer science chronic kidney disease
Pictured, l-r, are students Patrick Etim, Juan Figueroa, and Adithyan Karanathu Shibu. Absent from the photo is Dr. Jacob Levman.

The use of Artificial Intelligence (AI) for predicting whether a patient has chronic kidney disease is the subject of a new journal article published by computer science faculty Dr. Jacob Levman and StFX students Juan Figueroa, Patrick Etim and Adithyan Karanathu Shibu.

The manuscript is based on a course project completed in Dr. Levman’s biomedical computation course as part of StFX’s Post Baccalaureate degree in Artificial Intelligence (PBD in AI). It’s been accepted in the peer-reviewed journal Electronics in a special issue focused on biomedical applications of machine learning.

“Our manuscript produces particularly accurate AI diagnostic technology,” Dr. Levman says. It also demonstrated the value of clinical patient characteristics, like knowledge of whether a patient has an additional condition, like diabetes, informing the AI technology, he adds, and includes an additional analysis, called a cluster analysis, which demonstrated potential for the creation of future technology focused on characterizing the severity (or staging) of chronic kidney disease.

“Overall, this was an AI study with multiple points of value, helping to further our technological abilities applied to AI based diagnosis of chronic kidney disease.”

HUGE SUCCESS

Dr. Levman says the published article is particularly noteworthy as it is unusual for students to have their course project work, especially from an undergraduate course, published in a peer-reviewed journal.

“This represents a huge success for my students in undergraduate courses, a big success for the research software my lab has produced, and also a big success for the PBD in AI program.”

Student teams in Dr. Levman’s biomedical computation course last winter were tasked with developing an AI technology for a biomedical application. They used a research software package called df-analyze, which was created in Dr. Levman’s lab years ago that’s been through many iterations of improvements. The software has helped empower students to relatively easily create their own high quality AI technologies.

POTENTIAL TO HELP IMPROVE PATIENT CARE

“Juan, Patrick and Adithyan chose to focus on a diagnostic AI application for predicting whether a patient has chronic kidney disease. Technologies such as this have the potential to help improve the standard of patient care by helping to quickly and effectively detect the condition, potentially early on in its development. Early detection supports earlier treatments which are generally associated with improved patient outcomes.”

Dr. Levman says they had hoped to conduct an in depth analysis of AI technology for chronic kidney disease, hopefully achieving state-of-the-art predictive accuracy and the creation of technologies that are more easily explainable/relatable to clinical reality.

“Our approach achieved predictive diagnostic accuracy of 99.4 to 100 per cent, outperforming all studies in the peer reviewed literature focused on this same chronic kidney disease dataset.”

Some studies did not consider key patient characteristics, such as whether the patient has diabetes or hypertension, he said.

“My students noticed this when reviewing the literature and chose to perform this analysis with and without those patient characteristics. This demonstrated predictive potential from those patient clinical characteristics, as it clearly contributed to our state-of-the-art performance metrics. My students also elected to perform an additional type of AI analysis, called a clustering analysis, which demonstrates natural groupings in our dataset that appear to align to the severity of chronic kidney disease, thus our analysis also demonstrates potential for the development of future AI technologies for chronic kidney staging.”

Dr. Levman says the journal Electronics is an open access journal, which means that published papers are freely available for everyone to read, which is desirable for dissemination of scientific findings. However, open access journals charge publication fees to the submitting scientist, typically paid for from their research grants. Fortunately, he says, the journal invited him to contribute to this special issue and waived the open access fees.

“This is extremely exciting and is the culmination of tremendous effort in research, teaching and research-teaching integration,” Dr. Levman says.