Liver Tru-Cut — Primary Tumor Origin AI

AI-assisted inference of primary tumor origin from liver core biopsy whole-slide images

← Home

GitHub

https://github.com/sbalci/Liver-Tru-Cut-Tumor-Data-Preperation

• morphology-ai-research-liver-interface

https://github.com/sbalci/livertumorweb

WebPages

karaciger.patoloji.app

karaciger.patoloji.dev

Status

In the research wiki

• Liver Tru-Cut Cohort — the working data: pseudonymised slide archives and the clinical database, correctly held off git.
• Liver Tru-Cut — Tumour Origin AI — the project’s pointer page.
• Liver Working Tree — seven model environments standing up against one cohort — the clearest record anywhere here of what has actually been run.
• Weakly Supervised Learning — one label per slide is exactly this setting.
• Class Imbalance and Accuracy — demonstrated on this cohort rather than argued in the abstract: every zero-shot model scored below an always-predict-the-majority baseline.
• Foundation Models in Pathology — the encoders compared, and the finding that a pretrained slide-level aggregator did not replace a MIL head trained on local labels.
• hepatocyte-app — segmentation and morphometry running on the same cohort, and the reusable route to per-slide microns-per-pixel.