Reading Room Assistant (prototype)
A human-in-the-loop radiology reading-room prototype: prioritised worklist, imaging viewer with AI and radiologist annotation layers, prior-study comparison, AI-drafted impression and a sign-off step.
Important
Not a medical device. Never use for diagnosis.
The AI findings (regions, bounding boxes, confidence) are pre-authored demo data in www/data.js, standing in for a detection model that does not exist on the device. What is genuinely generated on the PC is the narrative: the image description and the draft impression (Phi Silica), grounded in that structured data.
How it works
flowchart LR D[seedStudies<br/>data.js] --> W[worklist<br/>computePriorityFromFindings] W --> V[viewer: zoom, window/level, compare prior] V --> BOX[renderBoxes<br/>AI amber / radiologist cyan] V -->|describeImage| ID[ImageDescriptionGenerator] BOX & ID --> DR[buildDraftPrompt → generate<br/>draft impression] DR --> SIGN[radiologist edits + signStudy<br/>audit trail] IMP[Import: .dcm / image] -->|parseDicom| V
Source files
| File | What it contains |
|---|---|
MainWindow.xaml.cs |
Ops ensureReady, generate, describeImage, parseDicom (fo-dicom: tags → age/sex/date, pixel data → preview PNG via ImageSharp). |
www/index.html |
UI: Worklist, Import Study, Dashboard, Responsible AI. |
www/app.js |
Worklist/priority logic, viewer (renderViewer, wireStageInteraction, applyImgFilter), side pane (findings, prior, impression, audit), prompts, sign-off, DICOM import. |
www/data.js |
Synthetic studies and the preset AI findings. |
www/images/*.png |
Synthetic images made by image-gen/generate_images.py. |
sample-import-data/ |
.dcm samples (made with pydicom, no patient information) and PNGs to try Import Study. |
Step-by-step deployment
Everything below is copy-paste PowerShell on a Windows 11 PC. Replace nothing: the repository owner is medyabi.
1. Check the PC and install the tools (once)
- Windows 11 build 22621 or later, x64 (
winver). For the AI features you need a Copilot+ PC (NPU); otherwise the app starts but reports the model as unavailable.
winget install --id Microsoft.DotNet.SDK.8
winget install --id Git.Git
winget install --id GitHub.cli2. Sign in to GitHub and clone (the repository is private, so the account must have access)
gh auth login
git clone https://github.com/medyabi/windows-local-ai-apps.git
cd windows-local-ai-apps3. Build the signed package (about 30 s to 2 min; the first build also downloads NuGet packages and needs internet)
.\scripts\build-app.ps1 -App ReadingRoomAssistantAppIt creates a self-signed certificate if needed and writes two files: dist\ReadingRoomAssistantApp\ReadingRoomAssistantApp_1.0.0.0_x64.msix and dist\ReadingRoomAssistantApp\ReadingRoomAssistantApp.cer. A warning about mspdbcmf.exe is harmless.
4. Trust the certificate (PowerShell as administrator, once per PC)
Import-Certificate -FilePath .\dist\ReadingRoomAssistantApp\ReadingRoomAssistantApp.cer -CertStoreLocation Cert:\LocalMachine\TrustedPeople5. Install
Add-AppxPackage -Path .\dist\ReadingRoomAssistantApp\ReadingRoomAssistantApp_1.0.0.0_x64.msix6. Launch it from the Start menu (Reading Room Assistant) or:
$p = Get-AppxPackage *ReadingRoomAssistant*
Start-Process "shell:AppsFolder\$($p.PackageFamilyName)!App"7. First launch
- The status pill shows the on-device model state. The first time, Windows may need to prepare the model (a few minutes, may need Windows Update).
- Prototype with synthetic studies; the AI findings are preset demo data. Import samples from
sample-import-data/.
8. If something fails
$p = Get-AppxPackage *ReadingRoomAssistant*
notepad "$env:LOCALAPPDATA\Packages\$($p.PackageFamilyName)\LocalState\bridge.log"Common errors and fixes are in docs/DEPLOYMENT.md.
9. Uninstall (also removes the app's data)
Get-AppxPackage *ReadingRoomAssistant* | Remove-AppxPackageTry it
- Worklist → open the chest study with a prior; switch between AI and radiologist layers, compare with the prior.
- Generate the draft impression, edit it, and sign.
- Import Study →
sample-import-data/chest-sample.dcm.
The page loads a Google Fonts stylesheet (cosmetic).