The Medical Machine Learning Lab (MMLL) is embedded within the Institute of Translational Psychiatry at the University of Münster. As a hub for Machine Learning and Predictive Analytics in Mental Health, we develop algorithms and build models to bridge the gap between machine-learning research and clinical application. For details of our research and a list of publications see our faculty webpage or www.photon-ai.com
Medical Machine Learning Lab (MMLL)
Theme and Objectives
The Medical Machine Learning Lab develops and applies machine-learning technology to bridge the gap between methods development and clinical application. To this end, we collaborate with numerous groups at the Faculty of Medicine (www.medizin.uni-muenster.de) and the University of Münster (www.uni-muenster.de) as well as national and international partners. The MMLL works in four broad areas.
Machine Learning Software Development – The Python-based PHOTON toolbox enables rapid machine learning model construction, optimization, evaluation, and deployment. It provides a high-level API automatizing the typical supervised machine learning workflow and integrates diverse toolboxes and frameworks including sci-kit learn and Tensorflow. Please visit www.photon-ai.com for details. For beginners, we provide a web-based graphical interface for machine learning code generation (PHOTON Wizard; www.wizard.photon-ai.com).
Machine Learning Algorithm Development – Addressing the two major challenges in medical machine learning today, we develop machine learning models for small and medium samples sizes. A particular focus is on medical decision making and risk evaluation under uncertainty using Bayesian Machine Learning. In this context, we actively develop machine learning solutions for high-dimensional structural and functional Magnetic Resonance Imaging (MRI) with a focus on 3d image and graph data.
Artificial Intelligence in Medicine – The AI revolution not only needs cutting-edge technological development, but it will be crucial to ensure clinical utility, safety and security. In this context, we actively develop best-practice guidelines and quality checks such as the AI Transparency Framework for Machine Learning in Medicine (https://osf.io/uzehj).
Benchmarks and Challenges – To focus attention on Machine Learning for Mental Health, we conduct the Predictive Analytics Competition (PAC), an annual Machine Learning challenges open for all (www.photon-ai.com/PAC).
- Supervised Machine Learning (Support Vector Machines, Gaussian Processes, Random Forests, etc.)
- (Bayesian) Deep Learning (Generative Adversarial Networks, Convolutional Neural Networks, Recurrent Neural Networks)
- Clustering, Bayesian Mixture Models
- Brain-behaviour relationships (Canonical Correlation Analysis, Partial Least Squares Regression)
- Python Machine Learning Software Development
- Python-based high-level API for rapid machine learning development (PHOTON)
- Graphical web tool for machine learning code generation (PHOTON Wizard)
- Development of Python toolbox PHOTON (www.photon-ai.com)
- Development of a browser-based graphical interface for machine learning code generation (wizard.photon-ai.com)
- Identification of biological subtypes of depression through multivariate clustering
- Unraveling brain-behaviour relationships using Canonical Correlation Analysis and Partial Least Squares Regression
- Bayesian Uncertainty Quantification in Multiple Sclerosis Prediction
Members of the Institute of Translational Psychiatry have acquired multiple intramural and extramural project funding and participate in several research initiatives funded by the DFG, BMBF, and EU. Selected funding sources include:
- DFG-Heisenberg-Professorship “Predictive Analytics in Mental Health” to Tim Hahn
- DFG research unit FOR2107 “Neurobiology of affective disorders” (co-speaker: U. Dannlowski), Project WP6 (see www.for2107.de)
- DFG grant “A Predictive Analytics Approach to the Optimization of Diagnosis, Treatment, and Ambulatory Management of Major Depressive Disorder and Bipolar Disorder”
- IMF grant “Phenotyping of Affective Disorders using Smartphone-Based Monitoring” (together with N. Opel)
- IZKF grant “NeuroML - Machine Learning Infrastructure for cerebrospinal fluid-based neurological discovery” (together with G. Meyer-zu Hörste)
Current selected publications
Leenings, R., Winter, N. R., Sarink, K., Ernsting, J., Jiang, X., Dannlowski, U., & Hahn, T. (2020). The PHOTON Wizard--Towards Educational Machine Learning Code Generators. arXiv preprint arXiv:2002.05432.
Leenings, R., Winter, N. R., Plagwitz, L., Holstein, V., Ernsting, J., Steenweg, J., ... & Hahn, T. (2020). PHOTON--A Python API for Rapid Machine Learning Model Development. arXiv preprint arXiv:2002.05426.
Flint, C., Cearns, M., Opel, N., Redlich, R., Mehler, D., Emden, D., ... & Hahn, T. (2019). Systematic Overestimation of Machine Learning Performance in Neuroimaging Studies of Depression. arXiv preprint arXiv:1912.06686.
Cearns, M., Hahn, T. & Baune, B.T. Recommendations and future directions for supervised machine learning in psychiatry. Translational Psychiatry 9, 271 (2019).
Lueken, U., & Hahn, T. (2019). Personalized Mental Health: Artificial Intelligence Technologies for Treatment Response Prediction in Anxiety Disorders. In B. Baune (Ed.), Personalized Psychiatry. Elsevier Science & Technology.
Domini, M., Monteiro, J. M., Pontil, M., Hahn, T., Fallgatter, A. J., John, S.-T., & Mourão-Miranda, J. (2019). Combining heterogeneous data sources for neuroimaging based diagnosis: re-weighting and selecting what is important. NeuroImage, 484311.
Cearns, M., Hahn, T., Clark, S., & Baune, B. T. (2019). Machine learning probability calibration for high-risk clinical decision-making. Australian & New Zealand Journal of Psychiatry.
Walter, M., Alizadeh, S., Jamalabadi, H., Lueken, U., Dannlowski, U., Walter, H., … Hahn, T., Dwyer, D. B. (2018, October 2). Translational machine learning for psychiatric neuroimaging. Progress in Neuro-Psychopharmacology and Biological Psychiatry.
Hahn, T., Nierenberg, A. & Whitfield-Gabrieli, S. Predictive analytics in mental health: applications, guidelines, challenges and perspectives. Molecular Psychiatry 22, 37–43 (2017).