FCAI SIG

AI for health

Breakthroughs in personalized medicine are expected to revolutionize medical care and have a strong long-lasting societal impact. AI will play a central role in this process and it is already making the first steps into the clinics. This page highlights how each of our research program interacts with healthcare research within FCAI and lists the groups currently involved.

Coordination: Professor Aleksei Tiulpin (University of Oulu), Professor Simo Särkkä (Aalto University) and Professor Mark van Gils (Tampere University)

Research Programs

Agile probabilistic AI provides explainable, verifiable, uncertainty-aware, and reliable tools for healthcare data analysis. The tools enable fast deployment of the healthcare AI solutions while providing rigorous methods for validation and uncertainty quantification.  These tools will benefit the public and private sector users of the AI by improving the outcomes of healthcare interventions, and will also benefit the medical researchers aiming to find solutions to fundamental challenges in diagnosing and treating human health.

Simulator-based inference enables the use of established knowledge in human health as part of personalized healthcare AI. Examples of knowledge that might be added to the AI simulations include models of human physiology and disease evolution. Including these models helps the AI to learn and generalize faster and better when using the smaller data sets that are typically available when treating an individual. The models are efficient and their reasoning is interpretable by a human reader. The models are even capable of learning causal relations by combining observations with experiments that the AI planned autonomously, which has a huge potential in modern medical research.

Next-generation data-efficient deep learning boosts the already ongoing success of deep learning in healthcare data analysis by providing AI methods that can cope with less training data while still sustaining the full power of deep learning. This ability to develop solutions with less initial data, especially in combination with simulator-based inference, allows for harnessing deep learning methods in many healthcare applications where the data is expensive or laborious to collect.

Privacy-preserving and secure AI provides AI solutions for healthcare, which preserve the privacy of healthcare data as well as the security and privacy of the developed AI systems. These kinds of solutions are particularly important in medical data analysis where ethical aspects as well as regulations such as GDPR have to be taken into account. Public sector healthcare data is an example of a valuable data source that cannot be utilized in AI solutions without proper privacy and security considerations.

Interactive AI ensures the natural interaction between the healthcare AI solutions, and the patients and healthcare providers. It allows for AI that patients and medical personnel feel comfortable to use, trustworthy, and worth using in practice. Developing patient and clinican trust in the system is essential for acceptance of AI solutions as day‑to‑day tools in medical practice.

People

The following groups already take part in FCAIs AI for Health research. If your group is already involved and needs listing here, please contact coordinating professors Simo Särkkä and Mark van Gils.

  • Sture Andersson, University of Helsinki

  • Jukka Corander, University of Helsinki, University of Oslo and Sanger Institute

    • Simulator-based inference, ABC, Population genomics, Vaccine development, Antibiotic resistance, Infectious disease epidemiology

  • Andreas Hauptmann, University of Oulu

    • Image reconstruction, Tomographic imaging, Medical image computing

  • Sampsa Hautaniemi, University of Helsinki

  • Mika Hilvo, VTT

  • Jaakko Hollmen, Aalto University

  • Antti Honkela, University of Helsinki

  • Leo Kärkkäinen, Nokia Bell Labs, Aalto University

    • Efficient, embedded, ways to utilize deep learning for medical data, Physical simulations to produce datasets for deep learning, GAN-based methods for improving transfer learning from simulations and for anonymizing datasets, Active learning

  • Kimmo Kaski, Aalto University

  • Samuel Kaski, Aalto University

  • Harri Lähdesmäki, Aalto University

    • Bioinformatics, Personalized medicine, Probabilistic modelling, Machine learning

  • Pekka Marttinen, Aalto University

    • Probabilistic modeling, Machine learning, Electronic health records, Statistical genetics, Mobile health

  • Ville Mustonen, University of Helsinki

  • Lauri Parkkonen, Aalto University

  • Matti Pirinen, University of Helsinki

    • Statistical genetics, Population genetics, Variable selection, Multivariate data analysis

  • Esa Pitkänen, University of Helsinki

  • Samuli Ripatti, University of Helsinki

  • Juho Rousu, Aalto University

    • Machine learning, Predicting structured data, Computational metabolomics, Personalized medicine, Systems pharmacology and pharmacogenomics, Systems and synthetic biology

  • Simo Saarakkala, Oulu University

  • Riitta Salmelin, Aalto University

  • Simo Särkkä, Aalto University – coordinator

    • Sensor and health informatics, Cardiovascular, AI multi-sensor medical diagnostics, AI for medical imaging, Biomedical sensing, Brain imaging

  • Arno Solin, Aalto University

  • Jing Tang, University of Helsinki

  • Aleksei Tiulpin, University of Oulu – coordinator

  • Mark van Gils, Tampere University – coordinator

  • Aki Vehtari, Aalto University

    • Bayesian data analysis, Probabilistic programming, Computational statistics

  • Ivan Vujaklija, Aalto University

    • Regression based estimation, pattern recognition and classification for neuroprosthesis, rehabilitation and prevention, and biomarkers.