Servizio Comunitario di Informazione in materia di Ricerca e Sviluppo - CORDIS

Final Report Summary - MIND (Modelling and Inference on brain Networks for Diagnosis)

Project objectives

The goal of this project was to develop techniques to use information about how parts of the brain are connected, so that clinical neuroscience can use this important data. Functional connectivity (the intrinsic organization of brain regions into networks when no task is performed) can be measured using a medical imaging (radiology) technique called magnetic resonance imaging (fMRI).
More specifically, we focused on the use of fMRI neurodegenerative diseases, by
a) Developing theoretical advances supported by empirical experiments on synthetic and real neuroimaging data of Multiple Sclerosis and Alzheimer Disease patients.
b) Exploring the practical limitations of the methods potentially impeding their widespread application.

Summary overview of results – engineering contributions

An important contribution was the development of new methods to compare brain connectivity between different imaging and non-imaging modalities. This allowed us to study the relationship between different biological processes happening in the brain. In particular, we have proposed graph-based statistical tests that can be used to test whether graphs representing modalities are related, and which parts of the graphs are similar. We have tested the approach on functional, structural, and morphological connectivity, as well as transcriptional similarity of gene expression in the brain. In this line of work, we have also developed and published guidelines for proper topological analysis of graphs representing functional connectivity data in clinical settings, so that spurious differences not due to the disease do not pollute the analysis.

A second important contribution was to develop techniques that allow imaging data to be combined across different acquisition sites. We have extensively studied the impact of having different acquisition hardware in large datasets pooling from multiple sites, and proposed two techniques showing that it is possible to reduce the impact of the site on functional MRI graph data in several ways, leading to improved diagnosis in an autism multi-site dataset.

The third major contribution was the release of the Pattern Recognition in Neuroimaging Toolkit (PRoNTo), in collaboration with several universities in Europe. This software allows easy access to state-of-the-art multivariate modeling methods for imaging data, which means that other researchers can save considerable time and avoid errors when they want to analyze their data. This toolkit is meant for predictive modeling, which is particularly well suited for computer-aided diagnosis applications – after training the computer, we can predict whether a new subject has a disease or is healthy.

Finally, we have also shown that classification (diagnosis) performance for functional connectivity graphs of depends critically on the choice of atlas (a way to divide brain images into regions) – even if the whole pipeline stays the same, changing the atlas can lead to dramatically different results. We have also proposed new correlation estimators for improved functional connectivity estimates, which yield much reduced bias, in particular with respect to the size of the brain regions involved.

Summary overview of results – scientific contributions

The main scientific contribution of this project was bringing functional connectivity closer to the underlying biology. We have shown that resting-state functional correspond to regions of high gene expression similarity. These findings were replicated in-vivo on adolescents, and show striking correspondence to axonal connectivity in the mouse. Building on this work, a translational track has been started, one on Alzheimer disease, and one on Fronto-temporal dementia. Here, initial results suggest that specific networks are related to specific genes, and that variations in these genes are related to disease status in neuropsychiatric diseases. By exploiting a neuroimaging endophenotype, this could represent an original pipeline for gene set discovery, dramatically reducing the need for the large numbers of patients usually recruited for genome-wide association studies. Further building on this work, we have shown that brain network topology alterations are related to cognitive performance in post-traumatic stress disorder (PTSD), and that this is related to expression of PTSD genes.

This project has also contributed to studying specific diseases and their relationship to functional connectivity, including Alzheimer disease and mild cognitive impairment. We have shown that Alzheimer disease and mild cognitive impairment patients have slower vascular reactivity (blood vessel dilation and contraction) than matched healthy subjects, and that this is related to cognitive performance. We have also built predictive models of disease evolution, to show when people might become clinically diagnosed with Alzheimer disease. Another disease that was studied is the genetic disorder 22q11 microdeletion syndrome, where several genes are missing in patients. People that suffer from this disease have a high chance of becoming schizophrenic at adolescence. We showed that these genetic modifications lead to very specific alterations of functional connectivity, and that these can be used to help classify schizophrenic from non-schizophrenic patients. This offers hope for better characterization of large-scale changes in brain organization due to the disease, and improved understanding of the disease.

Conclusions and Socio-economic impacts

By showing results in several different brain disorders, and in particular by linking functional imaging to genetics, the results of the project will bring functional magnetic resonance imaging (MRI) closer to being an accepted modality in clinical settings. We foresee that, similar to structural MRI which is now routinely used in clinical practice, functional imaging will play an increasing role in differential diagnosis, early diagnosis, and prognosis. More specifically, our results pave the way for integrating genetic information into an imaging workup, opening very promising avenues for precision medicine. This means that several brain diseases will become easier to distinguish at an earlier stage, resulting in improved outlooks for patients if early interventions or treatment, including lifestyle changes, are possible.

Informazioni correlate

Reported by