How do I use the Imaging Twin Sandbox? Think of it like a flight simulator for your outpatient imaging service. Start on the Setup tab and input your typical day length, arrivals per hour, no-show rate, number of scanners, number of readers, and average minutes for an exam and reading. These inputs represent the levers you control in real operations. Then, go to the Scenario Lab and select Run Simulation. The tool runs multiple Monte Carlo simulations in the browser and guides you to the Results tab. There, you’ll find KPI cards, two small queue-length graphs, and a table with one row for each replication. Make small, realistic adjustments one at a time. For example, increase the arrival rate during peak hours or add a reader, then rerun. Use Export CSV to review replications in Excel or Export JSON to save assumptions and summaries for a board presentation. Print or Save as PDF to capture a clear one-page summary. Keep the Methods tab accessible. It explains the logic, clarifies assumptions, and helps you brief colleagues without needing to look at the code.

The data shows how your system performs under load and identifies which step creates the true bottleneck. The average scan wait time and the percentage of scans waiting over 30 minutes reflect front-end access issues. The average read wait time and the percentage waiting more than 60 minutes reveal pressures on interpretive capacity. Time-to-Answer, especially its P90 tail, connects both stages and mirrors what patients and referring clinicians experience as total turnaround time. Throughput indicates how many studies you completed today, while Backlog at close shows work pushed into tomorrow. The queue charts illustrate the daily fluctuations of each stage, showing whether lines form early, stay at midday, or decrease with minor staffing changes. If adding a scanner reduces scan waits but leaves read waits high, reading is the bottleneck. If a small reduction in average exam time cuts both waits, workflow is the constraint. Use these patterns to justify targeted changes instead of broad expansion. Then, repeat the process with a different lever and verify that improvements are consistent across replications, not just in a single fortunate run.