Researchers German Cancer Research Center (DKFZ), Heidelberg University’s Medical Faculty, Heidelberg University Hospital developed AI system called Hetairos classify central nervous system (CNS) tumors using routinely prepared stained sections. research, published Nature Cancer, showed system classify 102 molecular tumor subtypes CNS cancers using digitized H&E slides, delivering diagnostic findings minutes rather than days or weeks often required molecular testing.
“ shows artificial intelligence is capable deriving molecular information directly routine sections thus fundamentally changing cancer diagnostics,” said lead author Darui Jin, PhD, postdoctoral fellow DKFZ.
CNS tumors encompass broad range diseases substantial molecular morphological diversity. methylation profiling is currently gold standard classifying many brain tumors, tests require specialized labs employing expensive analytical tools require sufficient amount tumor material, which difficult obtain some instances. these tests typically take two weeks return . reliance on molecular testing also created barriers some settings.
“Faster more widely accessible methods therefore needed,” researchers wrote. Existing alternatives such nanopore also require specialized instruments preparation. Routine H&E histopathology remains most widely available diagnostic material worldwide.
development Hetairos made possible advances computer vision digital pathology. Prior research established AI algorithms detect molecular features classify tumors molecularly defined categories. researchers noted previous research shown AI models estimate methylation signals standard H&E slides predict specific molecular alterations, “ artificial intelligence (AI)-based diagnostic solution H&E slides covers entire spectrum CNS tumors, currently only possible methylation testing, is still missing.”
develop Hetairos, team trained validated system using more than 11,000 digitized sections 9,606 treated 11 medical centers across four continents. Diagnoses used training were primarily established through methylation diagnostics. resulting model was designed classify tumors 102 molecular subtypes span nearly entire WHO classification spectrum CNS tumors.
important element system is it estimate confidence its predictions.
“Crucial Hetairos’s applicability across cohorts its realistic confidence estimates, which help judge its prediction accuracy,” researcher wrote. “Depending on cohort, Hetairos made high-confidence predictions 50–70% cases, those predictions were found correct nearly 90% instances.”
it become custom, system was also tested directly experts. Five board-certified neuropathologists reviewed 210 cases using only sections. Hetairos achieved accuracy rate 68%, compared 30% specialists. When three most likely diagnoses were considered, AI tool was 84% accurate versus 50% neuropathologists.
“ show modern AI systems now capable recognizing extremely subtle morphological patterns difficult even experienced specialists distinguish,” said Felix Sahm, project leader DKFZ.
prospective evaluation Hetairos practice analyzed 210 tumor alongside routine diagnostics without, its weren’t used influence management. While complete molecular testing required average 12 days, Hetairos generated approximately 12 minutes digitized slides were available. Including slide preparation scanning, findings often produced within 24 hours two days.
much speedier help clinicians initiate targeted treatments sooner guide additional testing, rather than replacing molecular testing, developers envision Hetairos triage decision-support tool.
“We developed Hetairos primarily tool support diagnostics,” Sahm said. “It is not intended replace molecular analyses, rather specifically complement accelerate them. technology make important contribution, particularly countries or regions limited resources, it is based on standard sections used worldwide.”
The post AI Tool Classifies 102 CNS Tumor Subtypes in Minutes appeared first on Inside Precision Medicine.