Techniques for Medical Imaging Analysis
Efficient, adaptable deep learning tools for robust image analysis.
This line of work focuses on pushing the robustness and efficiency of clinical image analysis through:
- Fine-tuning foundation models with low-rank adaptation (LoRA), adapters, and self-supervised learning (Finetune-SAM).
- Annotation-efficient learning Accelerating Volumetric Medical Image Annotation via Short-Long Memory SAM 2.
- Domain shifts evaluation and OOD detection Fréchet Radiomic Distance (FRD): A Versatile Metric for Comparing Medical Imaging Datasets.
- Test-time adaptation for robustness to distribution shift (InTEnt).
- Diffusion and generative models for synthesis and augmentation (Contour-diff).
- 2D to 3D reconstruction for reconstruct high-resolution 3D object from 2D images, like 2D Xrays (Fracture reconstruction), 2D MRIs (SuperMask).
- Registration algorithms for breast registration, like GuidedMorph.add
These techniques enhance both performance and usability in clinical environments.
Left: On-the-fly model adaptation. Right: Synthetic anatomy using generative models.