Hi there, I’m Eike.
I am a postdoctoral researcher at DTU Compute, in the section for Visual Computing. In my current research within the project Bias and Fairness in Medicine, I (together with my amazing collaborators) attempt to answer the following questions:
What does it mean for a data-driven health risk score model to be “fair”? How can we test whether a model is fair, and how can we design such a model?
Allow me to elaborate just a tiny bit. Health risk score models are likely to be used for resource prioritization in the healthcare system, for example by influencing who gets access to extra preventive care and who does not.1 As these are decisions that influence human livelihoods, we – as a society – would obviously want these decisions to be made “fairly”. But what does fairness even mean in this context?
- Should men and women (on average) get an identical amount of extra care?
- What if a disease has a higher prevalence in a certain group?
- What if we are better at identifying the high-risk patients in one group compared to another? Also, why would this happen in the first place?
- What if the data that are available are biased due to (historical or present-day) societal biases, such as poor patients having worse access to medical treatment compared to rich patients?
Once the philosophical question of which fairness definition to pursue is answered, various technical questions arise: how can the fairness of a model be quantified, and how can we actively build a model that is fair in the chosen sense?
Recent news
- I am co-organizing two exciting events on fair AI in medical imaging this fall, a workshop at MICCAI and a free, online symposium in November. Be sure to join in if you’re interested in these topics! We have an exciting line-up for both. (Sep ‘23)
- Our Cell Press Patterns perspective piece The Path Toward Equal Performance in Medical Machine Learning is out! Find a short summary here. (July ‘23)
- Presenting our work on risk score fairness and metric choices at the very inspiring FAccT’23 conference in Chicago! Also, we have published a preprint on the fairness of demographic invariance in medical imaging and a PNAS commentary on what to do when complete bias removal is not an option (May/June ‘23).
- Got invited to participate as a speaker in a wonderful Masterclass at the Lorentz center in Leiden on the clinical implementation of surface EMG measurements of the respiratory muscles (April ‘23).
- Went to MICCAI in Singapore to present our work on the impact of dataset group representation on MRI-based AD prediction performance (Sep), co-organized two events on fairness and responsibility in medical ML (1, 2, in Oct), and presented in a wonderful session on Biases in ML at the inaugural Danish Data Science conference (Nov ‘22).
- Two new journal papers with my previous group from Lübeck are out: Blind source separation of inspiration and expiration in respiratory sEMG signals and Model-based Estimation of Inspiratory Effort using Surface EMG. Also, I am now officially affiliated with the Pioneer centre for AI. (Jul ‘22)
- Our survey paper on responsible and regulatory ML for medicine got accepted and published, our MICCAI paper on feature robustness and sex differences in brain MRI got accepted, I wrote about climate change and AI, and we won funding by the DDSA for organizing a workshop on responsible ML for healthcare in autumn. (May/June ‘22)
- Had the honor of co-organizing (together with the amazing Laura Alessandretti) a workshop on Ethical, Secure, and Just AI at the opening event for the new Pioneer Centre for AI. We had an amazing list of speakers and panelists! (Mar ‘22)
- Our paper about surface EMG-based quantification of respiratory effort got published in Critical Care (Dec ‘21)
- Co-organizing a recurring seminar series on responsible AI now! Accessible via Zoom, everyone welcome. :-) (Dec ‘21)
- Started as a postdoc at DTU Compute with Aasa Feragen and Melanie Ganz (Sep ‘21)
Previously, I worked at the University of Lübeck, in the Institute for Electrical Engineering in Medicine. In a project executed together with the research unit of Dräger Medical, we worked hard to bring surface electromyographic monitoring of respiratory effort into clinical practice for improving mechanical ventilation. My research in this context spanned mathematical modeling, signal processing, parameter identification & statistical inference, all related to either surface electromyographic measurements, respiration, or both. See my previous publications for some of the work we did.
I am not saying that I believe this is a good idea. I’m just saying: it’s likely to happen, and, in fact, already happening. That does not mean that we, as a society, should not actively decide whether we want it to happen or not. ↩