Case Study: University Medical Center Phoenix accelerates COVID-19 AI reader study and identifies diagnostic CT imaging patterns with Flywheel

Radiologist Uses Flywheel to Accelerate Major COVID-19 AI Reader Study and Discover Diagnostic Imaging Patterns

University Medical Center Phoenix faced a diagnostic challenge early in the COVID-19 pandemic: PCR tests had limited sensitivity (~70%) and clinicians needed reliable imaging biomarkers to improve diagnosis and patient management. Radiologist Kathryn Olsen required a way to collect, de-identify and curate multicenter chest CTs and associated clinical data for a blinded reader study but lacked the IT infrastructure, so she turned to Flywheel and its AI development platform and radiology viewer.

Flywheel de‑identified and aggregated roughly 300 COVID‑19 chest CTs from 10 Colorado hospitals across four healthcare systems, integrated a custom data capture form, randomized cases, and assigned subjects to five expert radiologists with each scan scored by three readers using Flywheel’s secure centralized platform. The Flywheel-enabled workflow let Dr. Olsen manage the blinded multicenter study remotely, securely share de‑identified cohorts with Imbio for machine‑learning training, identify characteristic COVID‑19 CT patterns, and produce a reusable, measurable dataset (~300 subjects, multi‑reader scoring) to accelerate AI development and improve diagnostic and prognostic insights.

University Medical Center Phoenix

Daniel J. Rafter

Director of Product Management