Most Significant Contributions to Research

Many funding agencies require an applicant to identify their contributions to research. Here is a summary of my contributions.

Validation tools

I have created a digital brain phantom (Collins, Zijdenbos et al. 1998) [cited over 1000 times; source: Google Scholar] that was combined with MRI simulations. The resulting database of images was made publicly available (www.bic.mni.mcgill.ca/brainweb) and it has become a standard tool for validation and comparison of new image processing tools. This phantom was improved (Aubert-Broche,Evans, and D. L. Collins 2006) and extended to 20 different anatomies (Aubert-Broche, Griffin, Pike, Evans, Collins 2006).

Registration

Linear registation

I have developed an automatic technique for stereotaxic registration of MRI data (Collins, Neelin et al. 1994)[cited over 2400 times]. The procedure provides a framework for automated 3D analysis and is a basic image processing step of statistical analysis of functional data and voxel-based morphometry used by many research groups.

Non-linear registration

I have also developed an automatic technique for non-linear registration (Collins and Evans 1997; Robbins, Evans et al. 2004) to account for the residual local mismatch caused by second order differences between different subject and target anatomies that are not accounted for by linear registration. This is an important factor in cognitive functional data analysis (i.e., brain mapping) where it is necessary to compare results between subjects. The procedure has been used to estimate gross anatomical variability in human brain (Paus, Collins et al. 2001; Watkins, Paus et al. 2001) and has been used to characterize morphological differences between normal and diseased brain using voxel-based morphometry techniques in schizophrenia (Hulshoff Pol, Schnack et al. 2001; Hulshoff Pol, Schnack et al. 2004; Hulshoff Pol, Schnack et al. 2004; Hulshoff Pol, Schnack et al. 2005; van Haren, Hulshoff Pol et al. 2005), epilepsy (Bernasconi, Antel et al. 2001; Duchesne, Bernasconi et al. 2003; Bernasconi, Duchesne et al. 2004), MS (Prima, Ayache et al. 2002; Prima, Arnold et al. 2003) and many other diseases.

Segmentation

I developed ANIMAL (automatic nonlinear image matching and anatomical labeling). The technique uses non-linear registration to map regions from a pre-labeled atlas volume and onto the subject to achieve segmentation (Collins, Evans et al. 1995)[740+ citations](Collins and Evans 1997). This procedure has been used to customize a stereotaxic atlas for surgical planning (St-Jean, Sadikot et al. 1998) and evaluate surgical results (Atkinson, Collins et al. 2002). The segmentation tools have been improved by my students to include methods for active shape modeling that take into account the statistical properties of allowed shapes from a large set of example brain shapes (Duchesne, Pruessner et al. 2002; Hu 2005; Hu and Collins 2007) and to improve the quality of anatomical atlases for surgery (Chakravarty, Bertrand et al. 2003; Chakravarty, Bertrand et al. 2006). We combined ANIMAL with template selection and label fusion for hippocampus segmentation (Collins, 2010). Our more recent work involves non-local patch-based segmentation techniques (Fonov, 2011; Coupe 2010,2011,2012).

Medical Image Analysis

We have also developed tools for morphometric analysis using deformation-based analysis applied to the study of normal aging and differentiation with mild cognitive impairment and Alzheimer’s dementia (Duchesne, Bernasconi et al. 2005; Duchesne, Caroli et al. 2008), for aides to diagnosis and prognosis uses non-local means patch filters (Coupe, 2012) and cortical surface thickness (Eskildsen 2012). These techniques have been applied to the differentiation between normal subjects and patients with epilepsy (Duchesne, Bernasconi et al. 2002; Duchesne, Bernasconi et al. 2003) and the prediction of surgical outcome (Duchesne, Bernasconi et al. 2004). the patch-based techniques have been applied for diagnosis and prognosis in Alzheimer’s (Coupe 2011, 2012, 2015). In these studies, my students and I worked with clinical research groups to analyze data using our software analysis tools.

Image-guided surgery

We have developed a number of cognitive activation paradigms to map functionally important regions of the brain to assist in surgical planning (Amiez, Champod et al. 2005; Amiez, Kostopoulos et al. 2005; Amiez, Kostopoulos et al. 2005; Champod, Amiez et al. 2005; Champod, Amiez et al. 2005; Amiez, Kostopoulos et al. 2008). Finally, I have developed a new frameless stereotaxic technique that uses intra-operative ultrasound (Mercier, Lango et al. 2005) for improved guidance during surgery to account for brain shift (Arbel, Morandi et al. 2001; Arbel, Morandi et al. 2005). The procedure matches the intra-operative ultrasound to the pre-operative MRI data using non-linear registration. The result is a deformation field that can be used to warp the pre-operative data to fit the ultrasound, thus correcting for brain shift. Other procedures use blood vessels as landmarks to drive the registration(Reinertsen, Descoteaux et al. 2007; Reinertsen, Lindseth et al. 2007), and improved with simulated US (Arbel 2005, Mercier 2012a,b). In collaboration with Dr. Arbel (CIM, McGill) we recently implemented a GPU-based multi-modal MRI-US registration method(De Nigris, 2012a,b) that is based on alignment of gradient orientations of minimal uncertainty and achieves high accuracy in under one second.