Learning from limited data to
understand biological and medical images.
We connect microscopy, pathology, CT, and photoacoustic images with gene expression and clinical information. Our methods learn from limited labels and incomplete observations to address problems in medicine and the life sciences.
Seven research themes
We study biomedical image analysis and ways to learn from limited information.
Spatial transcriptomics
Linking pathology images to spatial gene expression, cell types, and gene relationships.
Explore the theme → 02 / CELL IMAGINGCell image analysis and generation
From cell detection, tracking, and mitosis analysis to microscopy image generation.
Explore the theme → 03 / VASCULAR IMAGINGVascular and multimodal medical imaging
Analyzing 3D CT and photoacoustic images, extracting vessels, and integrating clinical data.
Explore the theme → 04 / CLINICAL IMAGE ANALYSISPathology and endoscopic image analysis
Estimating pathology findings and disease severity from clinical images.
Explore the theme →Label-efficient learning
Learning with label proportions, partial labels, positive-unlabeled data, and multiple instances.
Explore the theme → 06 / ACTIVE ACQUISITIONActive learning and information acquisition
Choosing human or AI annotators, informative image pairs, and target-domain samples.
Explore the theme → 07 / ROBUST MEDICAL AIMedical AI under domain shift
Addressing changes in hospitals and imaging conditions across endoscopy, pathology, and cell images.
Explore the theme →Explore each research area and its publications.