Recently published
Self-hosted language models for real-world pathology coding
Clinical data science from routine care
With colleagues at IKIM, the group contributed to work on self-hosted language models for coding real-world pathology reports.
What problem are we trying to solve?
Can self-hosted large language models assign tumor ICD codes from real-world pathology reports with sufficient reliability for research and care-support use?
Why it matters
Manual coding of pathology is slow and uneven. Locally hosted models may reduce dependence on external cloud LLMs while remaining inspectable inside hospital infrastructure.
Approach
IKIM collaboration on self-hosted LLMs for ICD-O / tumor ICD coding (JCO CCI 2026; related preprint).