Our group has done significant work in the development of novel automated techniques for mapping the hippocampus and it’s subfields (Winterburn et al., 2013; Pipitone et al., 2014). We have developed a suite of detailed atlases that demonstrate the anatomy and architecture of this morphologically complex region of the brain. This work is currently being used to study healthy ageing and the hippocampus in the context of different neuropsychiatric disorders.

MAGeT : Multiple Automatically Generated Templates

The Multiple Automatically Generated Templates brain segmentation algorithm is a workhorse pipeline of CoBrALab. Implemented as a pythonic pipeline, integrated to take advantage of HPC clusters, MAGeTbrain is an antsRegistration-based segmentation pipeline, whose innovation is the "template" layer, which provides an intermediate step to transform input atlases towards the subject scans to be segmented. The registration phases contain a number of innovations, including an optimized multi-scale image pyramid, and masked registrations based on the labels to be propagated.


Multivariate methods for disease classification

Traditionally, scientists have used magnetic resonance imaging techniques to better quantify brain difference and neuroanatomical trends in healthy controls and disease groups. However, advances in machine learning algorithms that can handle vast quantities of data can allow us to use these data to automatically classify individuals into specific groups based solely on the neuroanatomical features that we derive. Our goal is build diagnosis algorithms that can perform the classification of individuals into disease groups based only on the information (using both native and derived measures) contained in magnetic resonance imaging and computationally derived indices of brain anatomy. We are currently developing projects that perform this type of classification in both schizophrenia and Alzheimer’s disease.

Understanding how the shape of neuroanatomical structures is relevant and implicated in healthy development and neuropsychiatric disorders

For decades neuroimaging scientists have been using brain volume as derived from magnetic resonance imaging data as a proxy for the integrity of a myriad of neuroanatomical structures. However, we have recently developed novel and sophisticated brain mapping techniques that estimate shape indices in different structures and we demonstrate how shape is changing in the brain in the absence of volume changes; this is consistent in our studies of normative development (Raznahan et al., 2014), neurodevelopmental disorders (Shaw et al., 2014), and addiction (Janes et al., 2014). We believe that these methods can elucidate subtle brain differences that may be invisible to more traditional volumetric measures.

NonNegative Matrix Factorization (NMF)

As the scale of available neuroimaging data rises, techniques which effectively identify relevant patterns become more valuable.  One of these possible techniques is non-negative matrix factorization (NMF), which has been shown to effectively represent complex data in an easy to digest manner.  In this project we use NMF to probe anatomical variability in certain brain regions such as the hippocampus.  In doing so we are able to identify regions of the brain that share similar characteristics across multiple MRI metrics and identify individuals which show varying properties in these regions.


Figure : The 4 component solution for both the left and right hippocampus, including 3-dimensional volumetric rendering 

              (Patel et al., NeuroImage, 2020)

Boundary Sharpness Coefficient (BSC)

The boundary sharpness coefficient (BSC) is a novel biomarker quantifying the transition between white matter and gray matter on magnetic resonance imaging data, more specifically on T1-weighted images. The BSC is defined as the growth parameter (i.e., slope) of the a sigmoid function fitted to several intensities running perpendicular to the gray-white matter boundary. More precisely, the procedure starts by generating 10 surfaces around the gray-white matter boundary (fig 1A). Then, the intensities (i.e., luminosity) of the image is sampled at on each surface (fig 1B) and a sigmoid curve is fitted to those intensities (fig 1C). A sharper transition between gray matter and white matter is reflected as a high BSC value, and a more gradual transition is reflected as a low BSC value (fig 1D). The BSC is thought to be mostly influenced by the myelin gradient around the gray-white matter boundary.


(Olafson et al., Biorvix, 2020)

RABIES: Rodent Automated Bold Improvement of EPI Sequences

Functional magnetic resonance imaging (fMRI) in rodents is an emerging field that allows for examination of brain networks in a preclinical setting. Currently, there are no standard approaches for obtaining fMRI measures of brain function, and the reliability and validity of such measures, which is critical for the interpretation of results, are only beginning to be systematically studied in rodents. To this end, we are investigating in the lab how the influence of anesthesia regimens, image processing strategies, as well as analysis techniques influence the results and interpretation of fMRI measures. This work will allow to make more informed decisions for experimental designs in rodent fMRI studies. We also contribute to reproducible and open science through the development of an open source automated image processing pipeline adapted for rodent fMRI ( RABIES ). The development of such software will be crucial to prevent failures of reproducibility across studies.


Figure : Processing diagram for the RABIES pipeline. (

Magnetization Transfer Imaging in Mice

Myelin concentration can be estimated non-invasively with a magnetic resonance imaging (MRI) measure known as the magnetization transfer ratio (MTR), yet the only implementation in the context of mouse studies has been with room-temperature MRI coils. We developed a technique for MTR mapping with cryogenically-cooled surface coils, to yield an improved signal-to-noise ratio. This technique overcomes ventral signal drop-off due to the coil shape by utilizing optimized sequence parameters and correcting for B1 field inhomogeneities. Surface cryogenic coil MTR maps corrected with this technique are more sensitive to changes in myelin content than the standard room-temperature coils.

Figure : Parameters optimization for mice magnetization transfer ratio (MTR) maps.