AI, (epi-)genomics, pathology |
Head of Tech @ Heidelberg Epignostix |
PhD @dkfz.bsky.social with Felix Sahm and @moritzgerstung.bsky.social | @imperialcollegeldn.bsky.social alum
Areeba Patel
Experts in Heidelberg have developed an AI system that can classify brain tumors with unprecedented accuracy using standard microscopic tissue sections. The system could accelerate the diagnosis of brain tumors worldwide. 🚀
t1p.de/fbfey
Tissue organization is fundamental to how tissues function - yet in cancer such spatial patterns can progressively break down. Ever wondered how?!
Check out our new study, where we investigate how lymphoma reshapes lymph node architecture 👀
3/n We trained and validated it on 11,000+ slides from 11 centres across 4 continents.
It recognises 102 tumour subtypes spanning the full paediatric and adult CNS spectrum, with calibrated confidence scores so you know when to trust a call.
4/n High confidence predictions in 50-70% of cases, accuracy 0.87 for top calls
Beat 5 board certified neuropathologists from over the world on H&E alone (0.68 vs 0.30)
In prospective routine use at Heidelberg: >90% agreement with final integrated diagnosis
1/n Excited to share our new paper in Nature Cancer 🦀
Meet Hetairos (Greek- “companion”): an AI model that predicts >100 methylation-defined CNS tumour subtypes directly from a routine H&E slide 🧠
State of the art diagnostics- from the slide that’s already on every pathologists’ desk 🧵
2/n The problem- methylation profiling gives the most granular CNS tumour classification (also recommended by WHO, NCCN, etc), but it’s expensive and takes ~2 weeks
H&E slides are everywhere, fast and cheap, but can’t deliver that detail on their own 🔬
Hetairos bridges that gap 🌉
6/n Why does it matter??
Hetairos can streamline diagnostics, streamline which cases need molecular testing and most importantly- bring up-to-date subtyping to centres where methylation profiling is simply not available.
It’s about making precision diagnostics accessible to all 🌏
5/n And it’s fast 💨
12 mins on consumer-grade hardware vs a >12 day turnaround for molecular testing ⌛️
Estimated cost- €1-2 per case 💸
A scanner and laptop are enough to get started
7/7 Huge thanks for my co-first authors Darui Jin and Artem Shmatko, and senior authors Felix Sahm and @moritzgerstung.bsky.social, and all collaborators ❤️
Read in Nature Cancer: www.nature.com/articles/s43...
Experts in Heidelberg have developed an AI system that can classify brain tumors with unprecedented accuracy using standard microscopic tissue sections. Using digitized standard stains, the system ide...
