Overview
- Effective connectivity and causality
- Brain parcellation
- Disease Heterogeneity Analysis
- Surgical devices segmentation and tracking
- Machine Learning and biological imaging
Research Projects
-
Effective connectivity and causality
The overarching goal of this project is to replace Pearson
correlations computed to measure functional brain connectivity and
structural covariances by more reliable and more informative alternatives.
In particular, I have improved the reliability
of my connectivity measures by employing recent covariance shrinkage methods;
I have investigated several tractable strategies to extract full brain effective connectivity
(where indirect connectivity has been factored out); and I am now exploring different
variants of Granger causality, which provides richer interpretations and more
complex models by informing us on the directionality of the functional connections.
more details
-
Brain parcellation
This project aims at the development of novel methods for the segmentation of the brain
into small connected regions of interests (parcels) exhibiting coherent properties.
In particular, when the BOLD signal acquired during resting-state fMRI scans is used for the
parcellation, the parcels obtained correspond to the functional units which are organizing the
brain function.
Starting with resting-state fMRI scans, I have gradually extended my approach to
handle other imaging modalities and to improve the speed of the computations.
In the near future, this project will benefit from the improvements granted by
the effective connectivity and the causality measures I am studying as part of the previous project.
more details
-
Disease Heterogeneity Analysis
Brain diseases are especially heterogeneous regarding clinical presentation and prognosis.
Moreover, the DSM-IV categories are suspected to produce diagnoses which are not associated
with the most relevant medications, which yields disappointing treatment responses.
These issues have recently led the NIMH to explore novel dimensional approaches, such as the Research Domain Criteria (RDoc).
On a broader scale, an emerging corpus of work is now focusing on the definition of neuroimaging
biomarkers for brain diseases, with the hope of better-predicting disease evolution and treatment response.
I took part in these efforts by designing semi-supervised methods able to discover reliable subgroups in patient populations.
This framework was used to test the presence of schizophrenia and neurodegenerative diseases subtypes.
more details
-
Surgical devices segmentation and tracking
The goal of this project was to segment and track the thin surgical devices used during cardiac angioplasty.
These tasks are particularly challenging because of the low signal-to-noise ratio of the fluoroscopic (real-time, low doses X-ray) images
acquired during the surgery.
more details
-
Machine Learning and biological imaging
Over the years, I have had the opportunity to process a broad variety of biomedical images, including
fluorescence images, fluoroscopic images, MRI, DTI, and resting-state fMRI scans, and
I implemented various Machine Learning methods to analyze these images. This section presents
the Machine Learning methods that I have specifically developed and implemented to analyze fluorescence images.
more details
