New publications

posted Jul 9, 2019, 7:15 AM by Juan Eugenio Iglesias

It is my pleasure to announce that we have recently had two new papers accepted.

In Dalca et al., accepted at MICCAI 2019, we propose a segmentation strategy that combines a conventional probabilistic atlas-based method with deep learning, enabling one to train a segmentation model for new MRI scans without the need for any manually segmented images. Segmentation only requires about 15 seconds at testing, and the code is freely available at

In Bocchetta et al., accepted at the Journal of Movement Disorders, we investigated brainstem involvement in atypical parkinson, using our brainstem segmentation tool. We managed to accurately quantify the differential involvement of brainstem structures in atypical parkinsonism.

You can find both articles under Publications.

Postdoc position available

posted May 1, 2019, 2:44 PM by Juan Eugenio Iglesias   [ updated May 1, 2019, 2:55 PM ]



We are looking for a postdoc in our ERC project, with two purposes. First, to improve the 3D histology reconstruction tools that we need to build an ultra-high resolution atlas of the brain using histology. And second, to collaborate in the development of segmentation techniques to apply the atlas to the automated analysis of in vivo MRI scans, which we plan to integrate within the widespread neuroimaging package FreeSurfer. Our research line is moving toward the integration of classic model-based algorithms with moder deep learning techniques (see for instance


The postdoc will join the UCL Centre for Medical Image Computing (CMIC), which will be soon co-located with the new UCL  AI  Centre on High Holborn. In addition, the postdoc will have the opportunity of visiting labs at MIT and the Martinos Center (Massachusetts General Hospital and Harvard Medical School) in Boston, where the PI holds affiliate / part-time appointments.

More information here:

Role of medial temporal subregions in primary progressive aphasia

posted Apr 5, 2019, 11:35 AM by Juan Eugenio Iglesias   [ updated Apr 5, 2019, 11:35 AM ]


We have a new paper accepted, on the role of medial temporal subregions in primary progressive aphasia. It turns out that there are interesting patterns in terms of staging and laterality in this disorder. Moreover, by looking at subregions of the hippocampus and the amygdala, it is possible to detect early changes that are not when looking at the whole structures.
Congrats on the paper, Martina & Jon!

IPMI paper on joint segmentation of structural and diffusion MR

posted Feb 27, 2019, 1:49 PM by Juan Eugenio Iglesias


We have an accepted IPMI submission on Bayesian segmentation of multi-modal diffusion / structural MRI data. We extend the traditional generative model of Bayesian segmentation with probabilistic atlases to include diffusion data (in addition to structural scans), which is beneficial to segment certain structures, such as the globus pallidus, the thalamus, and its nuclei. Due to its generative nature, the method can accomodate data acquired with a wide range of MR protocols, scanning platforms, etc. Our plan is to eventually release the methods, so they can be used in conjunction with our ex vivo atlases, particularly our recently presented thalamic atlas derived from histology. The paper can be found under Publications.

CTE: First, do not harm

posted Feb 19, 2019, 6:44 AM by Juan Eugenio Iglesias

A bunch of clinicians and researchers in traumatic brain injury and neurodegeneration (including me) have published a short correspondence in which we voice our concern by the tone in which the media have been reporting on of reporting chronic traumatic encephalopathy (CTE), e.g., as in this recent article in the New York Times [link]. Misleading information may have very negative consequences and we call for prudence and caution from the medical and research communities, as well as the media, when reporting on CTE. The complete communication can be found [HERE].

Paper on joint registration of tissue blocks

posted Dec 18, 2018, 8:49 AM by Juan Eugenio Iglesias


We have an accepted ISBI submission on joint registration of tissue blocks to a reference MRI scan. This is important when carrying out large-scale 3D histology reconstruction of the human brain with MRI as volumetric reference. To solve the problem, we propose a registration framework in which  blocks are simultaneously aligned to the MRI while exploiting the spatial constraints that they impose on each other, by means of a customized soft shape constraint - similarly to a jigsaw puzzle. We also propose  a hierarchical optimization strategy that takes advantage of prior knowledge on the slicing and blocking procedures. You can find the paper under Publications.

Paper on the effect of Fluorinert of fixed human brain tissue

posted Oct 3, 2018, 8:14 AM by Juan Eugenio Iglesias


It is my pleasure to announce that our paper “Effect of Fluorinert on the histological properties of formalin-fixed human brain tissue” has been accepted for publication in the Journal of Neuropathology & Experimental Neurology. Fluorinert (perfuluorocarbon) is an inexpensive option for minimizing susceptibility artefacts in ex vivo brain MRI scanning. In this brief report, we explore whether immersion in Fluorinert has an any impact on fixed human brain tissue or its histological analysis. Our qualitative and quantitative experiments using a convolutional neural network did not reveal any
discernible differences between samples that had been exposed to Fluorinert and samples that had not. Thanks to Shauna, Kate, Mohamed, Dave and Janice for all the hard work!

New paper on alignment of MRI and histological sections

posted Sep 13, 2018, 11:23 AM by Juan Eugenio Iglesias   [ updated Sep 13, 2018, 11:23 AM ]


Our paper "Joint registration and synthesis using a probabilistic model for alignment of MRI and histological sections" has been accepted for publication in Medical Image Analysis. It is known that cross-modality registration works better if one synthesizes one modality from the other and uses intra-modality metrics, rather than registering directly across modalities with e.g., mutual information. However, such an approach requires training data, ideally with perfect alignment. In this article, we present a method that solves for registration and synthesis simultaneously and directly test images, without need for training data. Using a probabilistic framework, synthesis and registration inform each other while accounting for the uncertainty in their estimates. Moreover, the framework can accommodate manually placed pairs of corresponding landmarks. The proposed approach is shown to outperform mutual information on synthetic images, and also on real data from BigBrain and the Allen Institute atlas. You can find the paper under Publications. Thanks to all the coauthors for the hard work!

Atlas of the human thalamus combining ex vivo MRI and histology

posted Aug 9, 2018, 8:11 AM by Juan Eugenio Iglesias

It is our pleasure to announce that our paper: "A probabilistic atlas of the human thalamic nuclei combining ex vivo MRI and histology" has been accepted for publication in NeuroImage. In this work, we present a probabilistic atlas of 26 nuclei of the human thalamus, which we have built by averaging manual delineations made on histological sections of the left and right thalami of 6 different post-mortem samples. The 3D structure of the histological data was recovered using a post-mortem MRI scan and blackface photographs taken during sectioning as references. The atlas shows good agreement with previous histological studies of the thalamus, and can be applied to automated segmentation of the nuclei in brain MRI scans of living people; we show in the paper that the atlas is able to detect differential thalamic effects in subjects with Alzheimer's disease, which are missed when the thalamus is analyzed as a whole. The atlas and segmentation tool are freely available as part of FreeSurfer (see Code). The paper can be found under Related Publications or Publications.

Multi-atlas for neonatal rabbit brain

posted Jun 8, 2018, 10:43 AM by Juan Eugenio Iglesias



We have a new paper in NeuroImage (first author: Sebastian Ferraris) describing a neonatal rabbit brain (multi-)atlas consisting of 12 multi-modal samples, with very precise manual annotations. Rabbit models are becoming increasingly popular in neurodevelopment studies, as they represent a convenient middle ground between small and large animals. The segmentations are are available at the following repository:
Congrats, Sebastiano, for the great paper!

1-10 of 68