CAIM Talk by Eugenio Iglesias: Ex vivo brain atlasing for generalizable segmentation: from classical techniques to neural networks
Speaker: Eugenio Iglesias (LEMON Lab, Harvard Medical School)
Time: 19.12.2025, 10am - 11am
Location: Anna Spiegel Research Building Seminar Room Level 3
Host: Georg Langs (CIR Lab)
Abstract: In this talk, I will describe our efforts to build high-resolution atlases of the human brain using MRI and histological data from donated specimens. I will begin with a brief introduction to Bayesian segmentation, which motivated our early work on probabilistic atlases of the hippocampus and amygdala using ex vivo MRI. I will then explain how incorporating histology allowed us to overcome key limitations of ex vivo MRI as we expanded to the thalamus, and how this shift introduced new challenges that we addressed with modern deep neural networks. This advances ultimately enabled us to scale our methods to the entire brain, and effort that culminated in NextBrain, a probabilistic atlas of the whole human brain derived from 3D-reconstructed histology and recently published in Nature. Next, I will discuss generalizable machine learning techniques developed in my group for analyzing brain MRI of any resolution or contrast without retraining, and how these methods have enabled robust, fast segmentation of 3D brain images in even the most challenging scenarios using NextBrain. I will conclude with an overview of ongoing related projects in our group, including applications to neuropathology, structural abnormalities, and beyond.
Short Bio: Juan Eugenio Iglesias is Associate Professor of Radiology at the Martinos Center for Biomedical Imaging (Massachusetts General Hospital and Harvard Medical School), where he directs the Laboratory for Ex Vivo Modeling of Neuroanatomy (LEMoN). He also has an affiliate appointment at the Massachusetts Institute of Technology (MIT). His research lies at the intersection of artificial intelligence and neuroimaging. Dr. Iglesias holds MSc degrees in Electrical and Telecommunication Engineering from the Royal Institute of Technology (KTH, Stockholm, Sweden) and the University of Seville, respectively. He completed his PhD in Biomedical Engineering at the University of California, Los Angeles (UCLA) in 2011. Dr. Iglesias has been the recipient of a Fulbright Science of Technology Award, a Marie Curie fellowship, a Starting Grant of the European Research Council, and several NIH grants.
Image from a recent paper by Eugenios group in Nature: Casamitjana et al. Nature 2025
This is part of the CAIM Talks series.
Machine Learning for Neuroimaging
Eugenio Iglesias team of the LEMON Lab at Harvard Medical School has made transformative contributions to brain atlas building, and turning atlases into a useful tool for brain image analysis, and quantification. Recently, his team published the first 100 micron resolution whole brain atlas with methods to link in-vivo- and ex-vivo imaging data (Casamitjana et al. Nature 2025). They have published SynthSeg to train robust domain independent segmentation models, SuperSynth, a foundation model for comprehensive brain image analysis, and recently SynthSR to tackle the mapping of multiple modalities. Overall these contributions scale quantitative precise neuroimage analysis to the heterogeneous reality of clinical imaging built into FreeSurfer.