New T-CAIREM AI in medicine courses explore the intersection of data science, machine learning
Alex Mariakakis, an assistant professor of computer science in the Faculty of Arts & Science, will teach the T-CAIREM courses.
Blame it on ChatGPT. In the past two years, the artificial intelligence (AI) tool has generated incredible public attention and seemingly revolutionized entire industries overnight. However, according to clinicians and researchers, AI’s biggest impact is still to come—and it just might be in healthcare.
To help students, researchers, and clinicians adjust to this brave new world, the University of Toronto’s Temerty Centre for AI Research and Education in Medicine (T-CAIREM) will launch three online courses in January 2025.
The courses will explore the basic building blocks of AI, namely data science and machine learning. They will also help familiarize learners at all levels with the ways AI tools such as electronic health records, automated diagnoses, patient triage, and chatbots improve workflows and patient care.
Alex Mariakakis, an assistant professor of computer science in the Faculty of Arts & Science, will teach the standalone courses. His research focuses on mobile health applications that apply machine learning to sensor data from smartphones and wearables to create new health interventions. As a result, Mariakakis works with data sources, such as vital signs, audio, and facial images used in AI health research. (One recent project, for example, included speech-analysis data to detect elevated respiratory symptoms.)
“Building these tools requires strong collaborations between experts in both healthcare and data science,” says Mariakakis. One of the biggest challenges in digital health occurs when practitioners from different fields have to work closely together without having a common language. “Participants in the upcoming T-CAIREM courses will acquire the vocabulary required to discuss the end-to-end AI health development pipeline.”
The three-course program originally focused on introductory programming. At the time, most people had little to no experience with AI but it quickly became apparent that the courses needed to accommodate beginners and learners curious about advanced topics. “I revamped the program in summer 2024,” says Mariakakis. “Based on the feedback from a pilot we ran, this latest iteration is a hit. I look forward to seeing it grow in the coming years.”
Mariakakis has made this program as accessible as possible. The first course, for example, is designed for people who have never written a computer program, while the second and third courses get into more advanced topics. Participants can take one course or all three in sequence and earn a certificate of completion.
Considering the high-tech nature of AI and medicine, will students have to invest in super-sophisticated computer equipment to enroll? “No, not at all!” says Mariakakis.
Course participants are not required to have specialized computer equipment. The lectures and homework assignments will be delivered using materials on Google Colab, a free cloud-based platform that runs Python code for data science tasks.
Each online course will cost $1500, which includes all required materials such as lecture notes and code examples. Lectures will be delivered as online videos on the University of Toronto’s Quercus learning platform.
As well, students can still ask questions and get help during the instructor’s scheduled office hours. Additionally, learners may post questions on discussion boards for each session that will be answered by Mariakakis or one of his teaching assistants.
“Anyone interested in data science could benefit from these courses, but there are definitely elements that cater to healthcare students and professionals,” he says.
Unlike the scores of publicly available online courses, tutorials, and videos explaining data science and machine learning, Mariakakis sees two major differences in the T-CAIREM courses.
“First, most entry-level machine learning courses focus on tabular datasets, but many people in healthcare want to work with other kinds of data like time-series sensor data, images, or audio. Second, most courses use datasets that are expected to yield accurate models, which typically isn’t the case for most real-world problems. We’ll talk about strategies for overcoming common challenges to improve model performance,” he says.
Mariakakis also notes that this program doesn’t explore advanced AI topics like deep learning, transformers, and LLMs.
“I know that many people are excited to learn about the latest innovations in AI, but it’s hard to get into them without a solid foundation,” he says. “My goal is to give learners the knowledge they need so that when they pursue these advanced topics, they can jump right in.”
