MICCAI paper on retrospective head motion estimation

posted May 16, 2017, 7:26 AM by Juan Eugenio Iglesias   [ updated May 16, 2017, 2:29 PM ]

Our paper "Retrospective head motion estimation in structural brain MRI with 3D CNNs" has been accepted at MICCAI 2017. 

In this study, we propose a supervised method to retrospectively detect whether voxels of a brain MRI scan show signs of being corrupted by motion or not. We can average the probability of motion across the voxels in a region of interest (e.g., the cerebral cortex) and analyze its impact on morphometric features (e.g., cortical thickness). We show that, when we factor in our estimate of motion in group studies, the conclusions of the analyses can be very different, particularly when one group is more prone to moving in the scanner than the other - in the paper, we illustrate this with a public, autism MRI dataset (ABIDE).

You can find a preprint of the manuscript under Publications.

Thanks Gari, Luis, Sara and Kepa for the hard work!

Computational atlas of the human amygdala and its nuclei

posted Apr 21, 2017, 1:52 PM by Juan Eugenio Iglesias   [ updated Apr 21, 2017, 1:53 PM ]

 Our paper "High-resolution magnetic resonance imaging reveals nuclei of the human amygdala: manual segmentation to automatic atlas" has been accepted for publication in NeuroImage. We used ultra-high resolution ex vivo MRI data to manually label the nuclei of the amygdala, and then used these manual segmentations to build a computational atlas of the nuclei - and surrounding structures. This atlas can be used to automatically segment the nuclei from in vivo brain MRI scans.  The atlas and associated segmentation code will be made publicly available as part of future versions of FreeSurfer. Thanks Zeynep, Dorit, Jean and the rest of coauthors, for the amazing work.

A preprint of the paper can be found here.

The role of hippocampal subfields in the atrophy process in Alzheimer's Disease (AD)

posted Mar 24, 2017, 5:51 AM by Juan Eugenio Iglesias   [ updated Mar 24, 2017, 5:54 AM ]


Our abstract on the role of hippocampal subfields in AD-related atrophy (first author: Marzi Scelsi) has been accepted for presentation at the Alzheimer's Association International Conference (AAIC), which will take place in London next July. Using our hippocampal subfield segmentation software, and the publicly available ADNI dataset, Marzia showed that some subfields are affected earlier than others in the disease, and concluded that subfield analysis may allow a better disease staging than the whole hippocampus alone. Congrats, Marzia (and Jonathan, Sebastien, Andre), and thanks for the hard work!

Postdoc position available

posted Mar 21, 2017, 1:25 PM by Juan Eugenio Iglesias   [ updated Mar 27, 2017, 8:39 AM ]

We are looking for a postdoc to work on my ERC project "Building Next-Generation Computational Tools for High Resolution Neuroimaging Studies" at the Translational Imaging Group (University College London). If you are interested, please apply through the link at the bottom of this page.
Further details on the position can be found here below.

Two year postdoctoral position in medical image analysis
Translational Imaging Group, University College London, United Kingdom.

The Translational Imaging Group (TIG) at University College London (UCL) is seeking to fill a postdoctoral position in medical image analysis within an exciting EU / ERC project. 

The postdoctoral researcher will be involved in the development of algorithms to recover the 3D structure of histological slices. The researcher will also help with the supervision of a doctoral student working on semi-automated / interactive segmentation of histological images within in the same project. The researcher will work in close collaboration with researchers at the UCL Brain Bank, who will be carrying out the histological processing of the (brain) specimens, and will also be using the interactive segmentation method to annotate large amounts of histological images.

These tasks are part of an EU/ERC project with the title "Building Next-Generation Computational Tools for High Resolution Neuroimaging Studies", awarded to Dr. J. Eugenio Iglesias. The project ultimately aims for creating an ultra-high resolution probabilistic atlas of the whole human brain using ex vivo data, as well as companion image analysis tools that apply the atlas to the automated segmentation of in vivo MRI scans. 

Applicant should have:

- A Ph.D. degree in (bio)medical image analysis or a related area (e.g., computer vision). 
- Strong mathematical and problem solving abilities.
- Strong programming skills (C++ and Matlab/Python essential).
- An established publication track record.
- Ideally, experience with medical image processing libraries and neuroimaging packages (ITK, VTK, SPM, FreeSurfer, FSL).
- Ideally, experience with analysis of histology data.
- Ideally, experience with image registration.

The postdoctoral researcher would be hired at grade 7 of the UCL salary scale ( We are looking for fill this position as soon as possible, but there is some flexibility with the dates. The duration of the contract would be two years, with possibility of further extension.

You can apply for the job here:

(Very) Preliminary results of THALAMODEL project

posted Feb 28, 2017, 9:21 AM by Juan Eugenio Iglesias


It is still very preliminary work, but we will be presenting an initial version of our thalamic atlas (for project THALAMODEL)  at the 2017 annual meeting of the Organisation for Human Brain Mapping (OHBM). See you in Vancouver!

IPMI paper on globally optimal coupled surfaces

posted Feb 20, 2017, 11:36 AM by Juan Eugenio Iglesias   [ updated Feb 20, 2017, 11:52 AM ]


My submission "Globally optimal coupled surfaces for semi-automatic segmentation of medical images" to IPMI has been recently accepted. This paper extends Leo Grady's PAMI paper "Minimal surfaces extend shortest path segmentation methods to 3D" to the multi-label scenario. In his PAMI paper, Leo presents a very clever extension of intelligent scissors to 3D, such that a user can for example manually segment two parallel slices of a 3D medical image, and let the algorithm fill in the rest by finding the globally optimal surface joining the two contours. Here we show that, with some modifications, we use the algorithm to compute N>2 optimal surfaces simultaneously without intersections or holes in between. 

Thank you very much, Leo, for all your idas and suggestions that made the paper possible.

New NeuroImage paper on mid-space-independent registration

posted Feb 20, 2017, 11:14 AM by Juan Eugenio Iglesias

Our paper "Mid-Space-Independent Deformable Image Registration" has been accepted for publication in NeuroImage. Using mid-spaces is a common way of making sure that registration algorithms are symmetric; however, the results depend on the mid-space definition. In this paper we present a formulation, inspired by statistical atlases, in which  the data term is independent of the mid-space, eliminating the need for anti-drift constraints. Thanks, Iman, Martin, Mert and Bruce, for the hard work!

A preprint of the paper can be found here.

New articles using hippocampal subfields

posted Feb 2, 2017, 4:17 AM by Juan Eugenio Iglesias   [ updated Feb 2, 2017, 4:18 AM ]

  Some exciting research using our hippocampal subfield module is being published! Hibar et al. (Nature Communications, in press) have presented a genome-wide association study (GWAS) of 33,536 individuals, in which they found that a locus within the MSRB3 gene shows evidence of a localized effect along the dentate gyrus, subiculum, CA1 and hippocampal fissure. Ho et al. (Neuropsychopharmacology, accepted) showed that, in a population of subjects with ultra-high-risk (UHR) for psychosis, a greater decline in the CA1 subfield happened in a group whose symptoms persisted and those who developed clinical psychosis, compared with UHR subjects who remitted and healthy controls.

In you are interested in analyzing your data with our subfield segmentation algorithm, the implementation is publicly available as part of the 6.0 version of FreeSurfer, which has been recently released.

New Neuroimage paper on MRI contrast adaptive segmentation of brain structures

posted Sep 6, 2016, 7:19 AM by Juan Eugenio Iglesias   [ updated Sep 6, 2016, 7:19 AM ]


Our paper "Fast and Sequence-Adaptive Whole-Brain Segmentation Using Parametric Bayesian Modeling" has been accepted in Neuroimage. Here we build up on well-established tissue segmentation algorithms to create a tool that segments brain structures from MRI scans, and which is robust against changes in MRI contrast thanks to its generative nature. This method is much faster than state-of-the-art multi-atlas (MAS) techniques, and the experiments show that:

1. It is not much worse than MAS when training and test data have perfectly matching intensity profiles (i.e., have been acquired on the same platform and with the same parameters).

2. Works better when the intensities are not matched.

You can find a preprint of the paper under Publications.

Congratulations, Oula and Koen, and thanks for the hard work!

New Neuroimage paper: Bayesian longitudinal segmentation of hippocampal subfields

posted Jul 7, 2016, 4:29 AM by Juan Eugenio Iglesias   [ updated Sep 9, 2016, 4:43 AM ]

We have extended our hippocampal subfield segmentation tool to the analysis of longitudinal MRI scans. Many of the questions we try to answer with neuroimaging are intrinsically longitudinal: effect of aging, progression of disease... We have just gotten a paper accepted in Neuroimage in which we present our new framework, which yields increased test-retest reliability and also higher sensitivity to group differences that the baseline cross-sectional approach. The proposed method rests on a generative model of longitudinal MRI data that relies on a subject-specific atlas. You can find the paper under Publications, and the code under Software->HIPPOCAMPAL-SUBFIELDS. Thanks Koen, Jean, Ricardo, Bruce and Martin for the hard work!

1-10 of 44