Case study: HistoPlus
Automatically detect cell types and tissue structures from histology data
Click to enlarge
5%
HistoPLUS outperforms SOTA with 5% better detection and 24% higher F1 classification, using 5x fewer parameters.
Converse directly with histological data
HistoPLUS is a state-of-the-art model for pan-cancer nuclei detection, segmentation and classification in H&E-stained pathology images.
- Characterize patient populations from tissue images
- Understand spatial composition of TME
- Identify biomarkers predictive of response
Predicted cell types and structures
HistoPLUS unlocks robust detection, segmentation & classification of 13 cell types, including understudied immune populations such as neutrophils and eosinophils.
- Cancer cell
- Lymphocytes
- Fibroblast
- Plasmocytes
- Macrophages
- Eosinophils
- Neutrophils
- Endothelial cells
- Red blood cells
- Epithelial cells (non-cancerous)
- Mitotic figures
- Apoptotic
- Smooth muscle cell / skeletal muscle cell
Development and performance
- HistoPLUS is built on top of the powerful H0-mini foundation model, developed in collaboration between Owkin and Bioptimus.
- HistoPLUS outperforms SOTA with 5% better detection and 24% higher F1 classification, using 5x fewer parameters.
- It generalizes across multiple cancer types.
Preprint and repositories
Read the preprint at: https://arxiv.org/abs/2508.09926
The model weights and an open implementation are available.
→ GitHub repository: https://github.com/owkin/histoplus
→ Model weights: https://huggingface.co/owkin/histoplus
Share to:
No items found.