Radiologist Skepticism
Responses ranged from enthusiastic early adopters to senior radiologists who found the overlay distracting and preferred to form their own assessment first. The technology was the easy part — adoption was not.
AI Diagnostic Tool: 3-Hospital Rollout for Chest X-Ray AI
How a regional hospital network deployed an FDA-cleared chest X-ray AI across three facilities in production — what the clinical workflow integration required, and what radiologists and clinicians actually experienced.
Get StartedAbout the Project
Deploying an FDA-cleared AI diagnostic tool in a real clinical environment is a different undertaking from a vendor demo or a research pilot. Production means the AI output appears in the radiologist's workflow for every eligible study, clinicians know how to act on it, and the integration runs reliably for thousands of studies a month.
This case study describes what that production deployment actually looked like across three hospitals in a regional network — the integration architecture, the clinical change management, and the outcomes the network measured over the first six months.
- Industry
- Healthcare / Radiology
- Business Type
- Regional 3-Hospital Network
- Core Offering
- FDA-Cleared Chest X-Ray AI in Production
- Build your idea
- Talk to our experts
The Clinical Problem Being Addressed
The network processed ~85,000 chest X-ray studies annually across its radiology departments, with unfilled radiologist positions creating workload pressure on the existing team. Time from acquisition to report was 4.2 hours for routine studies and 47 minutes for flagged urgent ones.
Leadership wanted to improve routine throughput, reliably surface critical findings such as pneumothorax, consolidation, and suspicious nodules promptly, and give radiologists a quality-check tool for high-volume reads.
The selected tool had FDA 510(k) clearance for pneumothorax, consolidation, pleural effusion, pulmonary nodules, and other chest pathologies. Leadership evaluated its published sensitivity and specificity as clinically meaningful, with particular attention to pneumothorax and consolidation — the highest-priority findings for timely detection.
Surface critical findings faster.
Support your radiologists.
Deploy AI that clinicians trust.
Talk to Our Team
The Integration Architecture
DICOM Routing from PACS to AI Engine
- Each PACS auto-sent chest studies to the AI endpoint at acquisition
- Targeted PA & lateral views from ED, inpatient & outpatient settings
- Excluded portable ICU ventilator-management views
AI Results Back into the Radiologist Workflow
- Findings returned to PACS as a DICOM Structured Report by the image
- High-confidence pneumothorax studies elevated to the top of the worklist
- Acuity, not acquisition time, drove reading priority
EHR Result Documentation
- Preliminary AI findings surfaced in clinician's EHR view before report
- Let ED physicians & hospitalists see findings awaiting the read
- Framed as preliminary — extra information, not a report substitute
Feedback Loop & Threshold Tuning
- Consistently dismissed AI findings reviewed in monthly QA meetings
- Judged whether dismissals reflected judgment, AI, or thresholds
- Confidence thresholds tuned from real clinical experience
The Clinical Change Management Challenge
Flexible Workflow, Not a Mandate
Rather than forcing one workflow, radiologists could reveal AI findings after their initial read, using the tool as a second-opinion check. Autonomy was preserved while findings still entered every read.
Clinician Education
ED physicians needed clear guidance on reading preliminary AI output in clinical context. The CMO sent a written communication to all ordering clinicians before findings appeared in the EHR.
Handling Disagreement
A simple protocol let radiologists document and dismiss AI findings with a brief note. Patterns of dismissal fed the monthly QA review, turning disagreement into deployment refinement.
What Made the Deployment Clinically Valuable
Over the first 90 days, the share of radiologists who preferred the AI overlay visible from the start of the read grew as the team built confidence in the tool — the deployment design earned trust rather than demanding it.
High-confidence pneumothorax studies jumped to the top of the worklist regardless of acquisition time — so the most urgent studies were read first.
Running on every study, the AI flagged findings not initially in the report; radiologists then confirmed and amended, or dismissed with justification. Leadership called this its most valuable function.
Radiologists reported lower cognitive load on routine studies and more confidence during high-volume sessions where fatigue could otherwise affect performance.
A flexible workflow let each radiologist choose when to see AI findings, which is why adoption grew steadily instead of meeting resistance.
The Results
31 min
Urgent ED Report Time
Down from 47 minutes
78%
Radiologists Report Positive Impact
At the six-month survey
85,000
Studies Processed Annually
Across all three facilities
3
Hospitals Live in Production
Network-wide rollout
Global presence
