Joakim Lindblad: Trustworthy AI-based decision support in cancer diagnostics

Modern AI-based techniques have the power to completely revolutionize health care. How can we improve the adoption and confident use of AI in everyday clinical practice? In this talk I will utilize our ongoing work towards large-scale implementation of AI-supported early detection of oral and oropharyngeal cancer as a source of a plethora of interesting challenges arising from this task. In particular I aim to share practical lessons learned from the multidisciplinary project, including discussions on different sources of data, multimodal information fusion, experiences of weakly and self supervised learning, creation of a collaborative annotation software, certainty estimation, explainability and interpretability.

Joakim Lindblad is a professor of computerised image processing at the Centre for Image Analysis at Uppsala University. He has extensive experience of both theoretical and applied development, use, and commercialization of image analysis methods and software. He is one of the inventors (Golf Magazine’s Innovator Award 2008) and main developer of the real time video tracking technology of Protracer AB (est. 2007), now TopGolf Sweden AB and has a lot of practical experience of turning ideas into products.

Following his PhD in Computerized Image Analysis at Uppsala in 2003 and a PostDoc at BC Cancer Research Centre, Vancouver, he held positions as Assistant professor at Uppsala and Novi Sad and Associate research professor at the Mathematical Institute in Belgrade. He is a senior member of the MIDA research group, associate editor of the SIAM Journal on Imaging Sciences, a board member of the Swedish Society for Automated Image Analysis (SSBA) and the Medtech Science & Innovation centre at Uppsala. His research interests are towards various aspects of Deep Learning for Image Analysis, including eXplainable AI, multimodal and self-supervised representation learning. 

He is creator of, and responsible for, several PhD and Master level courses on the topic of Deep Learning. He is currently PI for four research projects aiming to harness the power of modern data driven image analysis for practical use in biomedical and medical image data analysis.