IMAGINE Lab
At the Intelligent Medical Imaging and Image Processing (IMAGINE) Lab, we study image reconstruction, processing, and representation for medical and natural images by integrating generative artificial intelligence and physics-driven discriminative networks within computational imaging and inverse-problem frameworks, grounded in applied mathematics and physics, with a particular emphasis on MRI.
Annual lab gathering (05/10/2025).
News
Latest Updates
Paper accepted to CVPR 2026
Our latest work on consistency model-based inverse problem solvers has been accepted to the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR 2026).
New preprint on diffusion model representations
Our new preprint, “No Alignment Needed for Generation: Learning Linearly Separable Representations in Diffusion Models,” is now available.
Best paper finalist at IEEE CAMSAP 2025
Zero-shot Adaptive Diffusion Sampling (ZADS), our new approach to adaptive diffusion sampling in fast MRI, is a best paper finalist at IEEE CAMSAP 2025.
2 papers accepted to NeurIPS 2025 with one as a spotlight
Our latest work on accessible fast MRI reconstruction without raw k-space data and adaptive, time-aware algorithm unrolling for computational MRI has been accepted to NeurIPS 2025.
Prof. Akçakaya Named Inaugural Jim and Sara Anderson Chair
Prof. Akçakaya has been appointed as the inaugural Jim and Sara Anderson Chair.
Spotlight Oral at IEEE ICIP 2025
Our paper on sparsity-driven consistency for improved self-supervised MRI reconstruction has been accepted as a Spotlight Oral at the IEEE International Conference on Image Processing (ICIP) 2025.
Research
What we work on
Advanced MRI Reconstruction
& Enhancement
High-fidelity MRI reconstruction using physics-driven deep learning
- Novel image reconstruction and enhancement techniques
- Self-supervised and zero-shot PD-DL
- Time-embedded algorithm unrolling
- Physics-guided training for pTx pulse design
Generative AI
Improving generative model-based image generation and reconstruction
- Diffusion and consistency-based priors for reconstruction
- Adaptive likelihood-weighting
- Fast sampling & model distillation
- Improved sampling quality and speed
Highly-Accelerated & Motion-Robust Imaging
Highly accelerated and motion-resilient imaging across anatomies and acquisition strategies
- Ultra-high acceleration for cardiac MRI
- Simultaneous multislice (SMS) and fMRI
- Motion-aware reconstruction strategies
- Post-processing of medical imaging data
Recruiting
Interested in joining IMAGINE?
We are always interested in connecting with motivated students working at the intersection of inverse problems, MRI physics, and generative modeling.