HOTSPoT paves the way for scalable, explainable, and accessible AI in liver pathology.
Open science, clinical relevance, and technical excellence — all in one model.
Congrats to the entire team!
#DigitalPathology #OpenSource #AIinMedicine

Behind this innovation is a multidisciplinary, international team — but a special shoutout to Giorgio Cazzaniga, who led dataset creation, annotation, model design & implementation. A stellar example of clinician-engineering leadership. 🙌

Clinical validation? Yes. In 35 liver biopsies, HOTSPoT's automated tract count matched human observers (κ = 0.90) and correlated strongly with fibrosis stage (r = 0.87). A true step forward in reproducible pathology. 📈🩺

HOTSPoT isn't just a model — it's a tool. It's released in TorchScript format, integrates with QuPath, and includes a custom WSInfer pipeline for WSI-level inference. ⚙️
Code & model 👉 github.com/Gizmopath/HO...

Trained on 223 cases and tested across 5 international centers, HOTSPoT achieved Dice scores up to 0.92, showing minimal domain shift. That means: robust performance across scanners, stains, and disease contexts. 🌍📊

Liver histology is essential in diagnosing autoimmune and inflammatory diseases — but manual annotation is time-consuming and variable. HOTSPoT tackles this with automated, high-performance segmentation of portal tracts in H&E-stained slides. 🧠🧫