Machine Learning Infrastructure

Machine Learning Infrastructure

Refining, implementing, and making available state-of-the-art Machine Learning methods tailored to address current (clinical) research questions in a replicable and robust manner is of central importance.1 To this end, we have developed PHOTONAI as a high-level Application Programming Interface (API) to Machine Learning. It is designed to simplify and accelerate Machine Learning model development.2

It allows the user to easily access and combine algorithms from different ML toolboxes into custom algorithm sequences and is especially designed to automate the repetitive training, hyperparameter optimization and evaluation tasks. Importantly, the workflow ensures unbiased performance estimates while still allowing the user to fully customize the Machine Learning Analysis. PHOTONAI extends existing solutions with a novel pipeline implementation supporting more complex data streams, feature combinations, and algorithm selection. For an overview of the PHOTONAI ecosystem, see www.photon-ai.com .2,3

Importantly, PHOTONAI allows researchers to integrate heterogeneous data domains such as neuroimaging-, psychometric-, graph-, time-series, and clinical data. It enables Machine Learning architectures which combine data sources of arbitrary dimensionality and size to increase predictive performance.

We complement this software infrastructure with an online model evaluation platform which allows users to test model performance on data of different sites automatically.4 This publicly available machine learning model repository (www.photon-ai.com/repo ) provides an integrated infrastructure to evaluate model performance across sites at scale and thus enables direct and independent assessment of an ML model’s generalization performance and clinical utility. Already, the Enhancing Neuro Imaging Genetics Through Meta Analysis (ENIGMA) consortium, a large, international effort of more than 50 working groups, has adopted our platform to evaluate its Brain Age machine learning models.

Explore a preview of the PHOTONAI ecosystem

  1. Winter, N. R., Cearns, M., Clark, S. R., Leenings, R., Dannlowski, U., Baune, B. T., & Hahn, T. (2021). From multivariate methods to an AI ecosystem. Molecular psychiatry, 26(11), 6116-6120. ↩︎

  2. Leenings, R., Winter, N. R., Plagwitz, L., Holstein, V., Ernsting, J., Sarink, K., … & Hahn, T. (2021). PHOTONAI—A Python API for rapid machine learning model development. Plos one, 16(7), e0254062. ↩︎ ↩︎

  3. 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. ↩︎

  4. Leenings, R., Winter, N. R., Dannlowski, U., & Hahn, T. (2022). Recommendations for machine learning benchmarks in neuroimaging. Neuroimage, 257, 119298. ↩︎

Start your analysis with PHOTONAI