During the course we cover all relevant topics to understand and create AI systems for healthcare applications. Below you can get an impression of the content of the course and why these topics are covered.
All AI systems are written in programming languages. Python is the most commonly used language for AI algorithms. We believe that anyone that uses AI systems should at least have some idea what is happening under the hood. Therefore we offer everyone the opportunity to learn a bit of Python according to his or her own ambition. Our participants have broad backgrounds, from clinicians without any prior experience with Python, to dedicated software engineers. Some participants want to know all the details to be able to build complex AI systems all by themselves, others are happy to understand AI systems in more general terms. We try to facilitate these different ambitions at the start of the course by offering participants tutorials via Real Python at his or her own level of expertise.
During the machine learning lectures and practicals Joran Lokkerbol, Data Scientist at the Trimbos Institute, discusses the fundamentals of any AI system. Machine learning models are useful in situations where you have to predict or make decisions based on data existing of many variables with unknown relation to the prediction or decision to make. To understand how this works we cover topics such as understanding and preparing your data, the training and validation of models according to different performance metrics and the various machine learning methods that exist. Many of these fundamentals apply also to deep learning models, which are covered next in the course.
The deep learning lectures and practicals will look into a subset of the machine learning methods that use multiple ‘layers’ of computations, allowing them to learn more complex tasks. These models are very successful on tasks that were previously deemed impossible for computers, such as recognizing images or language. Within the Radboudumc we have great expertise on applying deep learning to medical images of all kinds of modalities. Bram van Ginneken, head of the Diagnostig Image Analysis Group, will provide lectures and practicals where you will learn how these models work, to what tasks these can be applied and what different methods there are.
All machine learning and deep learning models depend on a well prepared data set. This aspect of building a successful AI system is often marginalized, while in practice it takes up most of the time. To streamline this process we cover topics such as data collection and preparation. Together with the RTC Data stewardship we will cover the basics of data storage and usage according to the FAIR principles and introduce you to the Digital Research Environment of the Radboudumc. Rules regarding patient privacy will be covered by the privacy officers of the Radboudumc and ethical concerns will be discussed by Marianne Boenink, professor of ethics at the Radboudumc.
When dealing with large amounts of data it can be difficult to get a grasp of the most prominent features of your data. Visual representations can help you to quickly gain insight in your data and enables you to make appropriate decisions for the AI system you would like to create. Sara Sprinkhuizen will cover useful tools and tricks to visualize your data as clearly as possible based on scientific insights in how our brain processes information.
Once you have build a successful AI algorithm it still has to be implemented in practice. We will cover the deployment of AI systems together with PACMED, a company developing AI tools for hospitals. By taking us through their product pipeline, topics such as adjusting your AI system to your users, processing feedback and getting the right certifications are covered. AI systems fall under the Medical Device Regulation (MDR), which has recently been updated. Erik Gelderblom will cover the current regulation and what that means for using AI systems within the Radboudumc.
During this course you will work on a group project where you apply machine learning or deep learning to a real healthcare project. We use the CRISP-DM model (CRoss-Industry Standard Process for Data Mining) as a framework for working through the projects, and reporting on the project results. CRISP-DM structures the project work into the project phases of context understanding, data understanding, data preparation, modelling, evaluation and deployment. The projects will be brought in by the participants of the course where we will select the 5 most suitable projects where participants can work on in teams. We make sure that every team has an experienced programmer to enable the actual implementation of the AI system. For example projects, please vist the Example projects page.