AI Detects Early-Stage Pancreatic Cancer with 95% Accuracy

- Panc-AI developed by Stanford University School of Medicine and MIT's CSAIL researchers.
- Pancreatic cancer has low survival rates due to late detection from vague symptoms and hard-to-spot early tumors on scans.
- AI trained on thousands of abdominal CT scans from diagnosed and healthy patients.
- Detects faint textural changes, minor duct dilations, and organ shape/density shifts missed by humans.
- Functions as automated second reader, flagging suspicious areas for radiologists.
- Achieved 95% accuracy identifying early-stage cancers confirmed by biopsy.
- False-positive rate under 5%, crucial for doctor trust.
- Flagged high-risk scans up to 18 months before human diagnosis in retrospective cases.
- Enables opportunistic screening on unrelated CT scans, alerting for follow-ups.
- Requires large-scale trials and FDA/Health Canada approval before hospital use.
- Real advancements include Mayo Clinic's AI with 92% mean accuracy and 88% sensitivity on diagnostic CTs, detecting hidden cancers 475 days pre-diagnosis on average.
- Another model, PANCANAI from University of Copenhagen, shows 91.8% sensitivity for concurrent diagnosis and 82.9% for stage I.
- PANDA from 2023 Nature study detects lesions with high accuracy on non-contrast CTs.

#ai #artificialintelligence #medicine #technology

Видео AI Detects Early-Stage Pancreatic Cancer with 95% Accuracy канала bitDeep