AI PACS
Costa Rica and Canada
AI PACS is a review portal that runs an AI pre-read on uploaded medical images and gives radiologists a draft to check, edit and approve. It was built to support an existing PACS shared between Costa Rica and Canada.
- Sector
- Radiology across sites
- Setting
- A PACS shared between Costa Rica and Canada
- What it is
- An AI pre-read and review portal
- Roles
- Uploader, reviewer and administrator
The problem
When studies move between sites, they arrive with thin or inconsistent details. Radiologists spend time sorting and labelling before they can read, and urgent studies do not stand out.What we built
A portal where each upload is analysed for modality, anatomy, urgency and image quality. A reviewer approves or corrects the draft, every edit is logged against the AI's answer, and the images open in an OHIF viewer from an Orthanc PACS.About the project
The AI is a pre-read assistant, not a diagnostic authority. Human override is built in, and every correction is recorded.
What it does
What AI PACS does today, feature by feature.
Upload one file or a batch
Images are uploaded singly or in batches, and frames can be picked out of a video.
An AI pre-read
Each upload is analysed for modality, anatomy, urgency and image quality, with a confidence level.
The radiologist decides
A reviewer approves, edits or rejects the draft. The AI is a pre-read assistant, not a diagnostic authority.
Every correction is logged
Each human edit is stored against what the AI said, which gives the team an audit trail.
Instructions under version control
Administrators edit and version the instructions the AI follows, and each version is benchmarked against reference images.
A dashboard for the team
Counts and charts show what has been uploaded, reviewed and approved.
How it was built
The stages the project went through, in order.
- 1
Design the reviewer's workflow
Work out what a reviewer needs to see and decide for each study.
- 2
Build the review screens
Design and build the upload, study and review views.
- 3
Connect DICOM storage and the viewer
Link the portal to the PACS server and to the image viewer.
- 4
Add roles and the audit trail
Separate what uploaders, reviewers and administrators can do, and log every edit.
- 5
Deploy and refine
Release it and adjust it with feedback from the people reviewing.
How it fits in
What it connects to, who can use it and where it runs.
- Connects to
- An Orthanc PACS server for storage and an OHIF viewer for the images.
- Patient details
- Studies carry a de-identified patient ID, and the AI is instructed never to include patient identifiers.
- Access
- Sign-in is required, with separate roles for uploaders, reviewers and administrators.
- Who decides
- The reviewer. Human override is built into every step.
Last updated
Sponsored by Novo Nordisk