At a glance
- Client
- A multi-site diagnostic imaging networkName withheld for confidentiality
- Industry
- Healthcare
- Service
- Artificial Intelligence
- Scope
- Model development, PACS integration and clinical rollout
The challenge
Scan volumes at the network had grown far faster than its radiology team. Every study, routine or critical, entered the same first-in, first-out worklist, so a scan showing a bleed or a collapsed lung could wait behind dozens of normal chest X-rays.
Reporting delays were climbing, referring doctors were chasing results by phone, and radiologists were spending part of every shift scrolling through studies that needed no urgent attention.
Our approach
Map the reporting workflow
We sat with radiologists and technicians to see how studies moved from scanner to report, and agreed which findings counted as urgent.
Build and validate the models
We trained computer vision models on de-identified historical studies and validated them against radiologist reads before they touched a live case.
Fit into the existing tools
Instead of a new screen, results appear as priority flags inside the PACS worklist radiologists already use.
Roll out with clinicians in the loop
The system ran silently alongside the team first, then went live site by site with radiologists reviewing every flag.
The solution
Each incoming scan is pre-screened in the background. Studies with suspected critical findings move to the top of the worklist with the region of concern highlighted, while routine studies keep their normal place.
The model never replaces a radiologist's read. It changes the order in which scans are read, so the cases where minutes matter are seen first.
Results
- Urgent studies reach a radiologist first instead of waiting in a single queue
- Radiologists see suspected findings highlighted inside their existing PACS
- Referring doctors get critical results sooner, with fewer follow-up calls
- Every flag is logged for clinical audit and ongoing model monitoring
Technology used
- Python
- PyTorch
- DICOM
- PACS integration
- MLOps monitoring
Facing a similar challenge?
Tell us what you are working on. We will walk you through how we would approach it, with no obligation.