NeuroMiner Model Library
The last decade has witnessed the advent of machine learning techniques in the fields of clinical neuroscience. These methods hold the great potential to facilitate the translation of research findings into clinical applications following the principles of precision medicine. However they have been challenged by the lack of model transparency, external validation, and hence the lack of reproducibility. Furthermore, for most predictive models proposed for clinical neuroscience applications, it remains unclear whether they offer an edge over the current state-of-the-art.
To address these issues, the Section for Neurodiagnostic Applications headed by Prof. Nikolaos Koutsouleris at the Ludwig-Maximilian-University Munich decided to start a online model library that aims at facilitating the fully independent and external validation of predictive models in clinical neuroscience and mental health research. In its current form, the library is set up to host models which have been described in peer-reviewed companion publications, which are described on the respective model subpages. The models have been developed and cross-validated by researchers using
NeuroMiner: an open-source machine learning software available on Github.
Through a free account, users can access the library and enter data for prediction in a case-by-case fashion. This opens the possibility for external model validation and use of the models in experimental clinical settings, such as risk-stratified clinical trials. Please contact us here if you are interested in the use of our models for these research purposes.
The model library is in a pilot stage both in terms of its computational resources and its capacity to generate predictions using a variety of input data. Future versions of the library will support the use of neuroimaging data for prediction, as well as the use of machine learning software beyond NeuroMiner, such as TensorFlow or Scikit-Learn.