The Predictive Clinical Neuroscience group at the Donders Institute and RadboudUMC aims to develop statistical and machine learning techniques to make predictions relevant to brain disorders and to understand their underlying neurobiology on the basis of neuroimaging data. We focus both on supervised techniques for predicting clinical variables as well as unsupervised approaches for stratifying clinical groups on the basis of the underlying biology. Specific methodological techniques of interest include:

  • Bayesian non-parameteric methods for pattern classification and regression (e.g. Gaussian processes)
  • Spatial statistical methods (e.g. continuous stochastic processes, point processes)
  • Kernel and regularization methods for high-dimensional data
  • Multivariate statistical methods for finding associations between datasets (e.g. canonical correlation analysis)
  • Markov chain Monte Carlo methods for inference in probabilistic models
  • Multi-output, multi-task and structured output learning methods
  • Neural network and deep learning models

While the methods we develop have applications in many clinical settings, we have a particularly strong focus on psychiatric and neurological disorders.