Research
Code and data
We release code and trained models for most of our methods, so that others can reproduce and build on our work.
Physics-driven MRI and CT reconstruction
Accelerated MRI collects fewer k-space samples to shorten scans, and the missing information must be recovered during reconstruction. We develop deep learning methods that combine learned image priors with the physics of the imaging operator, so that reconstructions stay consistent with the acquired data.
Our recent work covers diffusion priors that adapt to each test scan, diffusion bridges constrained in the Fourier domain, zero-shot reconstruction without fully-sampled training data, and state-space models that unroll physics-driven updates. We also carry these ideas over to low-dose CT denoising.
Selected papers
- Learning Fourier-Constrained Diffusion Bridges for MRI ReconstructionIEEE Transactions on Medical Imaging, 2026Code
- Physics-Driven Autoregressive State Space Models for Medical Image ReconstructionIEEE Transactions on Medical Imaging, 2026Code
- Adaptive Diffusion Priors for Accelerated MRI ReconstructionMedical Image Analysis, 2023Code
- Unsupervised MRI Reconstruction via Zero-Shot Learned Adversarial TransformersIEEE Transactions on Medical Imaging, 2022Code
- DenoMamba: A fused state-space model for low-dose CT denoisingIEEE Journal of Biomedical and Health Informatics, 2025Code
- A Tutorial on MRI Reconstruction: From Modern Methods to Clinical ImplicationsIEEE Transactions on Biomedical Engineering, 2025Code
Multi-modal medical image synthesis
Multi-contrast MRI and multi-modal protocols give complementary views of the same anatomy, but some scans are often missing because of limited scan time, cost or patient tolerance. We develop methods that synthesize the missing contrasts or modalities from the available ones.
Our models range from conditional GANs and vision transformers to adversarial diffusion models, recursive diffusion bridges and selective state-space models. Several of them learn from unpaired or undersampled data, which lowers the need for curated training sets.
Selected papers
- Image synthesis in multi-contrast MRI with conditional generative adversarial networksIEEE Transaction on Medical Imaging, 2019Code
- ResViT: Residual Vision Transformers for Multi-Modal Medical Image SynthesisIEEE Transactions on Medical Imaging, 2022Code
- Unsupervised Medical Image Translation with Adversarial Diffusion ModelsIEEE Transactions on Medical Imaging, 2023Code
- Self-consistent recursive diffusion bridge for medical image translationMedical Image Analysis, 2025Code
- I2I-Mamba: Multi-modal medical image synthesis via selective state space modelingIEEE Transactions on Biomedical Engineering, 2026Code
- Semi-Supervised Learning of MRI Synthesis without Fully-Sampled Ground TruthsIEEE Transactions on Medical Imaging, 2022Code
Federated and privacy-preserving learning
Medical imaging data are spread across institutions and are hard to share because of patient privacy. We develop federated learning methods that train models collaboratively while the data stay at each site.
Our work focuses on the differences between sites. It includes generative image priors shared across institutions for MRI reconstruction, and personalized models that handle different sets of contrasts at each site for MRI synthesis.
Selected papers
- Federated Learning of Generative Image Priors for MRI ReconstructionIEEE Transactions on Medical Imaging, 2023Code
- One Model to Unite Them All: Personalized Federated Learning of Multi-Contrast MRI SynthesisMedical Image Analysis, 2024Code
Vision-language models and image analysis
We develop models that link medical images with clinical text and diagnostic labels. This includes methods that align medical vision-language models more closely with clinically meaningful findings, and transformer architectures for multi-label disease classification and content-based image retrieval.
Selected papers
- Meta-entity driven triplet mining for aligning medical vision-language modelsIEEE Journal of Biomedical and Health Informatics, 2026Code
- HydraViT: Adaptive multi-branch transformer for multi-label disease classification from Chest X-ray imagesBiomedical Signal Processing and Control, 2025
- Content-Based Medical Image Retreival with Opponent Class Adaptive Margin LossInformation Sciences, 2023
Magnetic particle imaging and imaging systems
Magnetic particle imaging (MPI) maps the distribution of magnetic nanoparticles with high sensitivity. Its image quality depends on a calibrated system matrix that is slow to measure. We develop deep learning methods that speed up calibration and reconstruct images with learned consistency.
The same principles carry over to other systems that depend on calibration, such as focal plane array imaging.
Selected papers
- TranSMS: Transformers for Super-Resolution Calibration in Magnetic Particle ImagingIEEE Transactions on Medical Imaging, 2022Code
- DEQ-MPI: A Deep Equilbrium Reconstruction with Learned Consistency for Magnetic Particle ImagingIEEE Transactions on Medical Imaging, 2024Code
- CalibFPA: A Focal Plane Array Imaging System based on Online Deep-Learning CalibrationIEEE Transactions on Computational Imaging, 2024
- Fast system calibration with coded calibration scenes for magnetic particle imagingIEEE Transaction on Medical Imaging, 2019
Functional brain imaging and computational neuroscience
We use functional MRI and machine learning to study how the human brain represents visual, auditory and linguistic information during natural behavior, and how attention reshapes these representations across the cortex.
On the methods side, we develop models for analyzing fMRI time series, including window-based transformers, graph neural networks that improve temporal sensitivity, and diffusion-based counterfactual explanations. Clinical collaborations include fMRI neurofeedback and connectivity studies in Parkinson's disease.
Datasets from our neuroscience experiments can be explored in the Brain Viewer.
Selected papers
- BolT: Fused Window Transformers for fMRI Time Series AnalysisMedical Image Analysis, 2023Code
- A Plug-In Graph Neural Network to Boost Temporal Sensitivity in fMRI AnalysisIEEE Journal of Biomedical and Health Informatics, 2024Code
- DreaMR: Diffusion-driven Counterfactual Explanation for Functional MRIIEEE Transactions on Medical Imaging, 2024Code
- Attention during natural vision warps semantic representation across the human brainNature Neuroscience, 2013
- Task-Dependent Warping of Semantic Representations During Search for Visual Action CategoriesJournal of Neuroscience, 2022
Earlier projects
Compressed sensing for accelerated MRI
Recently developed compressed sensing (CS) theory has opened up great avenues for accelerating MRI exams. The CS theory foresees that a compressible image can be sensed from a significantly smaller number of data samples than the Nyquist rate. To recover an image from these reduced set of samples, a linear transform of the image should be well represented by few coefficients. Furthermore, missing data samples should introduce aliasing artifacts with noise-like incoherent structure. MRI images exhibit similar statistics to natural images, and they are often highly compressible. However, the optimal sparsifying transforms and data sampling patterns for distinct body regions are not fully known.
We have ongoing efforts to apply CS to various MRI applications including MR angiography, MRI stem-cell tracking, and neuroimaging. We are developing novel automated parameter tuning algorithms that find dataset-specific reconstruction parameters to maximize image quality. These algorithms will empower practical use of CS reconstructions in clinical environments where manual tuning is impossible.
Steady-state acquisition methods for rapid MRI
MRI is a powerful modality that depicts the morphology and function of biological tissues noninvasively. Despite its increasingly important role in both clinical and research environments, image quality and spatiotemporal resolution in many MRI applications are limited by prohibitively long scan times. Thus, there has been a growing need for accelerated MRI methods.
To overcome the speed limitations of conventional MRI, we are developing a class of rapid pulse sequences based on steady-state free-precession (SSFP) imaging. SSFP sequences offer high signal-to-noise ratios, but can yield suboptimal tissue contrast and image artifacts due to signal loss. We tailor intelligent reconstruction algorithms to improve tissue contrast and artifact suppression in SSFP images to clinical standards.
Neural representation of visual objects and actions
Previous studies of visual category representation have identified a limited number of regions in the human brain that are highly responsive for specific categories. Well-known examples of these specialized regions include the fusiform face area (for faces) and the extrastriate body area (for bodies). These results have led to the hypothesis that category representation is mediated by highly localized and dedicated regions. However, humans can recognize thousands of unique visual object and action categories in natural scenes. It is therefore unlikely that a distinct brain region can be dedicated to each category.
To reveal the details of category representation across the entire brain, we are using natural movie stimuli that contain thousands of object and action categories. We first record brain activity with functional MRI while participants passively view these movies. We then employ a powerful voxel-wise modeling framework to quantitatively describe the selectivity of each voxel to each category. Our results imply that many brain regions represent information about a broader set of categories than previously assumed. We are currently examining the details of category representation across both visual and non-visual brain regions.
Attentional modulation of neural representations
Although real-world scenes are cluttered with many different objects, humans are extremely adept at finding target objects in natural environments and shifting their attention rapidly between different targets. The precise neural mechanisms that mediate this remarkable ability are yet unknown. Previous studies have reported relatively simple mechanisms that increase the quality of brain activity evoked by attended objects, without affecting the way that information is represented in each brain region. Yet, because there are a limited number of cortical neurons, it seems unlikely that all brain regions retain fixed representations irrespective of behavioral demands.
To study the nature of neural representations during visual search, we use complex natural movies and a powerful computational modeling technique to describe the relationship between visual information and brain activity. Our preliminary results suggest a novel attentional mechanism that reallocates neural resources across the brain to maximize sensitivity for the target and improve target detection under demanding conditions. You can explore our datasets here. We are currently studying the generality and flexibility of this mechanism for vision and for other sensory modalities.
