Get your hands dirty: Upcoming T-CAIREM courses help learners develop Data Science and Machine Learning skills
By
Asees Sandhu
We recently caught up with Alex Mariakakis, Assistant Professor in the University of Toronto’s computer science department, and the instructor of several T-CAIREM AI in medicine upskilling courses. Mariakakis also runs the Computational Health and Interaction (CHAI) lab, which explores how data from smartphones and wearable tech can improve people’s health. With that background, he’s well-qualified to teach the upcoming T-CAIREM courses to help learners from different disciplines get up to speed with data science for health.
What inspired you and T-CAIREM to launch the three-part Computing in Medicine Program?
My group’s research revolves around digital health and wearables, so I’m talking with various people in healthcare all the time. Along the way, I’ve picked up some of the language clinicians use and the considerations they find important. Knowing this really helps make my research as useful as possible to them. My primary motivation behind teaching these courses is to teach healthcare professionals about the language data scientists use and the considerations that we navigate in an approachable manner. Teaching them how to build machine learning models in Python helps everyone understand the strengths and weaknesses of coding.
For participants new to coding or AI, how do you make these technical concepts approachable and engaging?
I try to make lectures more engaging by using Jupyter Notebooks, which allows programmers to mix formatted text, images, and code all in the same interactive environment. In Jupyter, while I’m discussing different concepts, learners can play with code and see them in action.
Why are hands-on, real-world datasets such a powerful way to learn about AI in healthcare?
When I learned about machine learning, it was always with curated datasets that resulted in accurate machine learning models, as long as you did something reasonable. But when I applied those skills in my PhD research, things became much harder. My datasets didn’t come in the tidy packages I saw in my courses because they had all sorts of gaps and outliers. I had to translate what I learned about working with tabular datasets to other forms of data, such as images and time-varying signals, and that often required drawing on concepts from other courses. My goal with these courses is to talk through some of these challenges to make sure that learners don’t face the same struggles that I did.
Beyond technical proficiency, how do you hope this course changes the way participants think about technology’s role in patient care and innovation?
I really hope that people come away with an appreciation for the volume and diversity of data required to train a robust model. You can have the fanciest, most sophisticated machine learning model in the world, but if you don’t have enough training data, it probably won’t work very well in the real world and may even develop biases within the dataset. Let’s say you were building a model to detect skin cancer from photographs of skin lesions, but your dataset has only one example of a person with darker skin, and their lesion was benign. How can you be confident that your model will detect a malignant skin lesion on another person with darker skin? Efforts to produce large, comprehensive datasets have accelerated innovations in machine learning, and healthcare professionals will continue to play a significant role in developing new patient care technologies.
Your work at the CHAI Lab focuses on making healthcare more human through technology. In a field driven by data and algorithms, how do you keep the human element at the heart of innovation?
Part of my research agenda lies in the subfield of computer science called human-computer interaction (HCI), which covers everything from the design of user interfaces to the study of how people behave when using technology. In even simpler terms, it’s about designing technology that’s easy to learn and meets users' needs.
Looking ahead, how do you see AI transforming healthcare?
There are all sorts of new and exciting ways that people are thinking about adding AI into their clinical workflows: diagnostic decision aids, patient triage, transplant prioritization, remote patient monitoring, chatbots, and more! I’m really excited about remote patient monitoring. All of the inexpensive medical equipment we’re seeing come to market is making it easier for patients to capture basic information from the comfort of their own homes, reducing unnecessary clinic visits while extending their healthcare coverage.
What advice would you give professionals eager to be part of the integration of AI in healthcare?
For those who are super excited about integrating AI in healthcare, there’s no better way of learning than getting your hands dirty. You can start by building something simple, then gradually adding to it over time. And that’s exactly what we’re going to do in this program!
