Diagnostic modelling of multimodal neuroimaging data in mental disorders

• Applicant: Privatdozent Dr. Benedikt Sundermann
• Subject Area: Clinical Neurology; Neurosurgery and Neuroradiology ; Nuclear Medicine, Radiotherapy, Radiobiology
• Term: from 2016 to 2017
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 311084090

Today, mental and several systemic brain disorders are mainly diagnosed with a symptom-based approach. This entails limitations to differentiate between biologically different disorders with similar clinical presentation and to predict the course of disease in individuals. Thus, there is currently a high desire to complement the diagnostic spectrum with diagnostic biomarkers based on neuroimaging data, particularly from magnetic resonance imaging (MRI). Such functional, morphometric and microstructural imaging techniques of the human brain are well-established scientific tools to study neural correlates of mental disorders. Despite the scientific success of these mainly univariate or unimodal analyses no standard approach is sufficiently robust to serve as a reliable diagnostic test in individual patients. There have recently been promising reports about applications of multivariate classification techniques (in contrast to standard univariate techniques) to build diagnostic models based on e.g. functional MRI data. These analysis methods aim at jointly using information from multiple brain regions by specific discriminative analysis methods to boost the diagnostic power of inherently noisy measures. However, a majority of these studies only used a single imaging modality. It is assumed that using concordant, as well as complementary, information from multiple modalities can improve diagnostic power of such dedicated diagnostic models. It is a particular challenge to combine multimodal imaging data (and potentially clinical information) into one single diagnostic model, as data typically differ considerably regarding structure and distribution. Purposes of this project are (i) to implement an explicitly diagnostic modelling approach for joint analyses of multimodal MRI datasets in mental disorders that can optionally be extended by further sources of information, (ii) to optimize and validate this approach in existing large-scale datasets and (iii) to arrange its application to further diagnostic questions when sufficient diagnostic accuracies can be achieved in ii.This one-year project uses state-of-the-art and experimental data processing strategies for functional, volumetric and diffusion-weighted MRI data as well as an innovative strategy to combine these datasets: linked independent component analysis (linked ICA). These methods were developed at the host institution. They will be combined with multivariate classification techniques to form diagnostic models. The resulting data analysis pipeline will be optimized and tested based on data from a German cohort study on major depression (1392 subjects) and on data from a study on ageing with a focus on dementia. A further focus will be to prepare the application of the optimized data analysis strategy to further diagnostic questions in future studies by ensuring the extensibility and usability of the software implementation with clinical application in routine care as a long-term goal.

• DFG Programme: Research Fellowships
• International Connection: United Kingdom
• Host: Professorin Gwenaëlle Douaud, Ph.D.