Aleksei Tiulpin, PhD

Assistant Professor
Research Unit of Health Sciences and Technology
Faculty of Medicine
University of Oulu, Finland

Aapistie 5A
90220 Oulu

Email: firstname [dot] lastname [at] oulu [dot] fi

Google Scholar

About me

I am an Assistant Professor at the University of Oulu, and I am also a member of the European Lab for Learning and Intelligent Systems (ELLIS). The focus of my research is on Intelligent Medical Systems, and I develop new machine learning methods for medical applications. Earlier, I was a post-doctoral fellow at the Finnish Center for Artificial Intelligence and Aalto University working with Prof. Samuel Kaski and Prof. Simo Särkkä. Before that, I was a post-doctoral fellow at KU Leuven, working with Prof. dr. Matthew Blaschko.

I earned my PhD during 2017-2020 (graduated with distinction) at the Faculty of Medicine, University of Oulu. My PhD thesis was nominated the best doctoral thesis of the year 2020. My advisors were Prof. Simo Saarakkala, PhD (primary), Dr. Jérôme Thevenot, PhD, and Prof. Esa Rahtu.

I was awarded a title of docent (i.e. habilitation) in Machine Learning for Medical Imaging on 01.02.2022.

Intelligent Medical Systems Research Group

Motivation. My group develops new AI methods for medical applications. The research that we are conducting is motivated by the following challenges that we either already face with medical AI, or will be facing in very near future:

  • Today's AI does not treat patients, doctors, and hospital personnel as a part of the decision-making workflow.
  • Developers do not (and cannot) optimize the right healthcare metrics: patient outcomes, as well as diagnostic and treatment costs.
  • Methods that we have on the machine learning side have achieved success in tasks that are closer to automated data analysis and pattern recognition than to actual automatic decision-making.
So, how can we address the defined challenges? I envision that in order for patients to gain benefit from AI systems, we need to develop new methods on the Machine Learning side that

  • Are multimodal
  • Use as little labeled data as possible
  • Predict how uncertain they are in their predictions
  • Interact with humans and other AI systems
Vision. Powerful methods are good to have, but this is still not enough, as basic Machine Learning technology is merely an enabler. I believe that in order to see substantial improvements in health and care, we need to build new hospital processes, which are AI-tailored by design. The future AI-tailored hospitals (FAITH) will comprise AI systems designed for all major stakeholders in a hospital workflow (including the hospital itself), and those systems shall be seen as a new type of medical software. In practice, I foresee the following implementation of FAITH:

  • Patient’s AI: Personalized (to patients) diagnostic tools, and adaptive behavioral intervention systems.
  • Doctor’s AI: Personalized (to doctors and patients) prognostic and treatment recommendation tools.
  • Hospital’s AI: Productivity tools.

Group members. Currently, my group has the following members

  • Egor Panfilov, MSc, Postdoc
  • Huy Hoang Nguyen, MSc, Postdoc
  • Khanh Nguyen, MSc, Doctoral Researcher
  • Helinä Heino, MSc, Doctoral Researcher
  • Narasimharao Kowlagi, MSc, Doctoral Researcher
  • Terence McSweeney, MSc, Affiliated Doctoral Researcher
  • Santeri Rytky, MSc, Affiliated Doctoral Researcher
  • Zhisen Hu, MSc, Affiliated Doctoral Researcher
  • Trung Dang, BSc, Research Assistant
  • Petri Partanen, BSc, MSc student


My PhD thesis (best doctoral thesis of 2020 award): Download PDF.

The full list of my papers can be found on Google Scholar. The list of selected projects:

Nguyen, H. H., Blaschko, M. B., Saarakkala, S., & Tiulpin, A. (2023). Clinically-Inspired Multi-Agent Transformers for Disease Trajectory Forecasting from Multimodal Data. IEEE Transactions on Medical Imaging.
[Link] [Code]

Nguyen, K, Nguyen, H. H., & Tiulpin, A. (2021). AdaTriplet: Adaptive Gradient Triplet Loss with Automatic Margin Learning for Forensic Medical Image Matching. (MICCAI 2022; top-13% of all papers)
[Link] [Code]

Tiulpin A. & Blaschko M.B (2022). Greedy Bayesian Posterior Approximation with Deep Ensembles. Transactions on Machine Learning Research.
[Link] [Code]

Nguyen, H. H., Saarakkala, S., Blaschko, M. B., & Tiulpin, A. (2020). Semixup: In- and Out-of-Manifold Regularization for Deep Semi-Supervised Knee Osteoarthritis Severity Grading From Plain Radiographs. IEEE Transactions on Medical Imaging, 39(12), 4346–4356.
[Link] [Code]

Tiulpin, A., Melekhov, I., & Saarakkala, S. (2019). KNEEL: knee anatomical landmark localization using hourglass networks. In Proceedings of the IEEE International Conference on Computer Vision Workshops (pp. 352-361)
[Link] [Code]

Tiulpin, A., Klein, S., Bierma-Zeinstra, S. M., Thevenot, J., Rahtu, E., van Meurs, J., Oei, E. & Saarakkala, S. (2019). Multimodal machine learning-based knee osteoarthritis progression prediction from plain radiographs and clinical data. Scientific Reports, 9(1), 1-11.
[Link] [Code]

Tiulpin, A., Thevenot, J., Rahtu, E., Lehenkari, P., & Saarakkala, S. (2018). Automatic knee osteoarthritis diagnosis from plain radiographs: a deep learning-based approach. Scientific Reports, 8(1), 1-10.
[Link] [Code]