Resources
Case studies
Published validation studies and clinical-research deployments using Carina AI products.
-
CURATAI · University of Pennsylvania
University of Pennsylvania — defacing privacy vs. utility, head-to-head
Penn benchmarked four defacing algorithms — including CuratAI's — across CT, MRI, PET, and dose maps from 185 brain and head-and-neck patients. CuratAI delivered the best privacy–utility trade-off, preserving organ volumes with no measurable changes across all structures.
-
CURATAI · STAR Registry (multi-site)
STAR neurovascular registry — 87.7% accuracy on a 12 GB consumer GPU
Validation of CuratAI's local LLM extraction against expert REDCap ground truth on the Stroke Thrombectomy and Aneurysm Registry. Local 7B model reaches 87.7% accuracy on 170 fields per patient — within 3.2 points of a cloud baseline, with no PHI leaving the institution.
-
INTCONTOUR · Mayo Clinic Arizona · Radiation Oncology
Mayo Clinic Arizona — when the vendor model isn't enough
Vendor auto-contouring models underperformed on Mayo Arizona's prostate protocol (iodinated spacers, rectal balloons). The team retrained INTContour on 100 local cases. Blinded review by 6 GU radiation oncologists found AI contours rated ≥ manual in ≥50% of cases across every structure.