We develop AI models that turn sleep EEG, ECG, and oxygenation signals into interpretable markers of brain aging and glymphatic function. Cross-modal generative transformers connect nightly physiology with MRI-derived brain-clearance measures, enabling scalable and remote monitoring of neurodegenerative risk.

Through the ENIGMA Glymphatic Working Group, we are building harmonized lifespan reference models of perivascular spaces, free-water, DTI-ALPS/RaPPiD, and global BOLD–CSF coupling across international cohorts.

Head motion during MRI acquisition presents significant problems for subsequent neuroimaging analyses. In our most up-to-date work (Sun et al., ISMRM 2025), we aim to develop a Motion-Adaptive Diffusion Model (MADM) to correct motion artifacts in MR, improving image quality.  MADM is based on a diffusion model. Gaussian noise is added in the forward process, and a U-Net progressively denoises the images in reverse process. Our model was trained on the MR-ART dataset.

Diffusion Model

What we can do: A generative model used to   create images by simulation  the diffusion process. Key Idea: Start with random noise, and gradually transform it into a meaningful image. Forward Process: The model adds noise to the images over multiple steps, simulating a diffusion process. Reverse Process: The model learns to reverse the diffusion process, starting from noise and progressively generating a structured output via neural networks

Our deep-learning models estimate brain age at regional and network scales to reveal disease vulnerability, compensation, and recovery. In a multicohort ENIGMA study published in The Lancet Digital Health, contralesional neuroplasticity was associated with motor impairment after chronic stroke.

We integrate sleep quality, brain-age models, glymphatic MRI, amyloid/tau PET, and blood biomarkers to predict therapeutic response. Applications include CPAP and upper-airway stimulation for sleep apnea, stroke rehabilitation, and anti-amyloid treatment—toward precision neurology grounded in interpretable multimodal models.

White matter hyperintensities (WMHs) are abnormal signals within the white matter region on the human brain MRI and have been associated with aging processes, cognitive decline, and dementia. In the current study, we proposed a U-Net with multi-scale highlighting foregrounds (HF) for WMHs segmentation. Our method, U-Net with HF, is designed to improve the detection of the WMH voxels with partial volume effects. We evaluated the segmentation performance of the proposed approach using the Challenge training dataset. Up to date, the proposed method has achieved the best overall evaluation scores, the highest dice similarity index, and the best F1-score among 39 methods submitted on the WMH Segmentation Challenge that was initially hosted by MICCAI 2017 and is continuously accepting new challengers. The evaluation of the clinical utility showed that the WMH volume that was automatically computed using U-Net with HF was significantly associated with cognitive performance and improves the classification between cognitive normal and Alzheimer's disease subjects and between patients with mild cognitive impairment and those with Alzheimer's disease. The implementation of our proposed method is publicly available using Dockerhub (https://hub.docker.com/r/wmhchallenge/pgs).

Changes in brain morphology, such as cortical thinning are of great value for understanding the trajectory of brain aging and various neurodegenerative diseases. In this work, we employ a generative neural network variational autoencoder (VAE) that is conditional on age and is able to generate cortical thickness maps at various ages given an input cortical thickness map. To take into account the mesh topology in the model, we propose a loss function based on weighted adjacency to integrate the surface topography defined as edge connections with the cortical thickness mapped as vertices. Compared to traditional conditional VAE that did not use the surface topological information, our method better predict “future” cortical thickness maps, especially when the age gap become wider. Our model has the potential to predict the distinctive temporo-spatial pattern of individual cortical morphology in relation to aging and neurodegenerative diseases. (ISBI 2021)

This research aims to use innovate MR Imaging techniques to quantitatively characterize normal fetal / neonatal brain development as well as atypical development associated with prematurity. This will be accomplished using a large database of MR scans which have been obtained twice per baby, as well as using various deep learning methods and imaging processing techniques, which correct head motion artifacts, accurately segment brain structures, and quantify tissue characteristics and brain structural and functional network properties. Next, development of an online platform integrated with a big data-driven deep learning algorithm predicts the neurodevelopmental outcome of preterm babies and will ultimately personalize early diagnoses of motor, language or cognitive impairment, playing a significant role in planning customized rehabilitation for each survivor.

BACKGROUND:

Dramatic brain morphological changes occur throughout the third trimester of gestation. In this study, we investigated whether the graph convolutional network (GCN) that account for cortical morphometrics and cortical surface topology as a sparse graph can predict preterm neonatal brain age. Moreover, we evaluated whether the predicted brain age (PBA) is associated with postnatal abnormalities and neurodevelopmental outcome. 

METHODS:

577 T1 MRI scans of preterm neonates from two different datasets were analyzed; the NEOCIVET pipeline generated cortical surfaces and morphological features, which were then fed to the GCN to predict the brain age globally and regionally. The brain age index (BAI; PBA minus chronological age) was used to determine the relationships among preterm birth (i.e., birthweight and birth age), perinatal brain injuries, postnatal events/clinical conditions, BAI at postnatal scan, and neurodevelopmental scores at 30 months. 

RESULTS:

GCN-based age prediction of preterm neonates without brain lesions (mean absolute error [MAE]: 0.96 weeks) outperformed conventional machine learning methods using no topological information. Structural equation models (SEM) showed that BAI mediated the influence of preterm birth and postnatal clinical factors, but not perinatal brain injuries, on neurodevelopmental outcome at 30 months of age. Regional BAI of several frontal cortices correlated with cognitive abilities at 30 months whereas BAI of left Broca’s area was associated with language functional scores.

CONCLUSIONS:

GCN provide a clinically meaningful index in measuring brain age, localizing regional growth as it relates to postnatal factors, and predicting neurodevelopmental outcome.