ImagiVue — User Manual

Comprehensive guide to the medical image viewer

1. Getting Started

Browser Requirements

ImagiVue requires a browser with WebGPU support:

• Google Chrome 113 or later (recommended)
• Microsoft Edge 113 or later
• Safari 18 or later

WebGPU is required for GPU-accelerated rendering and AI segmentation. If your browser doesn't support it, you'll see a compatibility warning.

Loading Files

• Drag and drop — Drag files or folders directly onto the viewer
• File picker — Click "Open Files" in the bottom-left controls and select files or a directory
• Demo datasets — Click the demo icon in the file picker to browse sample datasets
• DICOMweb / PACS — Click the PACS button in the toolbar to search and load studies from a hospital PACS server
• URL launch — Open a study directly via URL parameters (see Section 8)

Supported File Formats

Standard image formats (JPEG, PNG, etc.) are loaded as non-medical images. They appear under a "Non-Medical Format" patient in the hierarchy and support basic viewing, annotation, and measurement tools.

2. Mouse Controls

The scroll wheel behavior can be changed between slice navigation and zoom using the scroll mode button in the bottom controls. Hold Shift while scrolling to temporarily invert the mode.

3. Keyboard Shortcuts

4. Window / Level Presets

Window/level adjusts image brightness and contrast. Use the preset dropdown or drag on the image:

5. Annotation Tools

Select a tool from the toolbar to begin annotating. Click on the image to place annotation points.

ROI Statistics

Region-based tools (Rectangle, Circle, Ellipse) display ROI statistics for the enclosed area:

• Mean — average pixel/HU value in the region
• Std Dev — standard deviation of values
• Min / Max — minimum and maximum values
• Unit — HU (Hounsfield Units) for CT, raw intensity for other modalities

Annotation Editing

Select an annotation to reveal interactive handles for editing:

• Move — Drag the center handle to reposition the annotation
• Resize — Drag corner or edge handles to change the size
• Rotate — Drag the rotation handle on ellipse annotations to change orientation
• All edits are tracked in the undo/redo history

Managing Annotations

• Click the annotations button (clipboard icon) in the top-right to open the annotations panel
• Click an annotation in the list to select and highlight it on the canvas
• Click the pencil icon to add a custom label or RadLex finding code to an annotation
• Click the copy icon to duplicate an annotation to an adjacent or specific slice
• Click the trash icon to delete an annotation
• Use Ctrl+Z / Cmd+Z to undo and Ctrl+Shift+Z / Ctrl+Y to redo any annotation action (create, delete, edit, move, label change)
• Use the Export dropdown to save annotations as JSON, DICOM SR, COCO JSON, NIfTI masks, or CSV
• Use the Clear dropdown to remove annotations for a series, study, or patient
• Annotations are automatically saved to IndexedDB and persist across sessions

Annotation Labels

Each annotation can have a custom text label and an optional RadLex finding code. Click the pencil icon next to an annotation to open the label editor:

• Type any custom text in the label field
• Select a standardized RadLex finding from the dropdown (e.g., Nodule, Mass, Calcification, Fracture)
• Labels appear on the canvas next to the annotation and in the annotations panel
• Label changes can be undone with Ctrl+Z

Copying Annotations

Annotations can be copied to other slices within the same series. Click the copy icon next to an annotation:

• Previous slice — Copy to the slice before the current one
• Next slice — Copy to the slice after the current one
• Custom slice — Enter a specific slice number to copy to

Copied annotations are independent of the original and can be edited separately.

Exporting Annotations

Use the Export... dropdown in the annotations panel:

• Export Series / Study / Patient / All — download annotations as a JSON file
• Export DICOM SR — export as a standard DICOM Structured Report (.dcm) file following the TID 1500 Measurement Report template, compatible with PACS systems
• Upload SR to PACS — send the DICOM SR directly to the active PACS endpoint via STOW-RS
• Export COCO JSON — export in COCO format for AI/ML training (research)
• Export NIfTI Mask — export segmentation masks as NIfTI volumes (research)
• Export CSV Summary — tabular summary of all annotations

You can also import previously exported JSON files back into the viewer.

AI Segmentation

AI segmentation runs entirely in-browser using ONNX Runtime Web. Available models:

• MedSAM — Medical image segmentation based on Meta's Segment Anything Model, optimized for medical images
• SAM-Med2D — 2D medical image segmentation variant
• LiteMedSAM — Lightweight version for faster inference
• MONAI Label — NVIDIA's medical imaging AI framework (server-assisted)
• Vista3D — 3D volumetric segmentation supporting 127 anatomical structures

Select the model in Settings (gear icon). Models are downloaded on first use (50-360 MB) and cached locally in IndexedDB.

6. AI Findings

When AI algorithms produce findings (detected regions, lesions, abnormalities), they appear in the Findings Panel:

• Findings are displayed with confidence scores and color-coded severity levels (critical, high, medium, low, info)
• Severity-colored bounding boxes appear on the image as overlays
• Accept / Reject / Defer — review each finding and mark your decision
• Filter findings by status (pending, accepted, rejected, deferred) or severity level
• Navigate — click a finding to center the viewport on it
• A statistics bar shows total, pending, accepted, and rejected counts

7. Pixel Readout

As you move the cursor over an image, pixel information is displayed at the bottom of the viewport:

• Single-frame grayscale: XY:[x, y] [value]
• Multi-frame grayscale: XYZ:[x, y, z] [value]
• Color images: XY:[x, y] [R, G, B](RGB)

For CT images, values are displayed in Hounsfield Units (HU).

8. PACS / DICOMweb Connectivity

ImagiVue can connect directly to hospital PACS systems via the DICOMweb standard. Click the PACS button in the top toolbar to open the DICOMweb panel.

Worklist Tab

Search for studies on the active PACS server:

• Enter a patient name (partial matches supported) and click Search
• Filter by modality and date range
• Results show patient name, study date, modality, and description
• Studies you've previously loaded show a "viewed" indicator
• Click a study to load it into the viewer — series and images are fetched on demand
• Pagination support for large result sets (50 per page)
• The active server name is shown above the search bar

Connections Tab

Manage DICOMweb endpoint connections:

• The Orthanc Demo (UCLouvain) server is pre-configured as the default
• Click + Add Connection to add a custom PACS endpoint
• Enter a name and DICOMweb base URL (e.g., https://your-pacs.hospital.org/dicom-web)
• Select authentication type: None, Basic (username/password), or Bearer token
• Use the Test button to verify connectivity before saving
• Select the active endpoint using the radio button — the worklist will use this server
• Custom connections are saved in your browser and persist across sessions

STOW-RS Upload

Annotations exported as DICOM Structured Reports can be uploaded directly to the active PACS endpoint using STOW-RS. Select "Upload SR to PACS" from the Export dropdown in the annotations panel.

URL Launch (EHR Integration)

Studies can be opened directly via URL parameters, enabling integration with Electronic Health Record (EHR) systems:

?studyUID=1.2.3...&wadoRsRoot=https://pacs.example.com/dicom-web&auth=basic&user=...&pass=...

• studyUID — the DICOM Study Instance UID to load
• wadoRsRoot — base URL of the WADO-RS endpoint
• endpoint — name for the connection
• auth — authentication type (basic or bearer)
• user / pass — credentials for Basic auth

Study History

Recently loaded studies from PACS are tracked in your browser (up to 500 entries) and shown with a "viewed" indicator in the worklist.

All DICOMweb traffic goes directly from your browser to the PACS server. No patient data passes through ImagiVue's servers. Your hospital IT may need to configure CORS headers on the PACS to allow browser access.

9. Grid Layout & Multi-Viewport

Use the grid layout selector in the bottom controls to display multiple series simultaneously. Available layouts include:

• 1x1 — Single viewport (default)
• 1x2, 2x1 — Two viewports (horizontal or vertical)
• 2x2 — Four viewports in a grid
• 2x3, 3x2 — Six viewports
• 3x3 — Nine viewports
• 4x4 and larger — Up to 24 viewports for advanced comparison

Click a viewport to make it active. Additional controls:

• Linked window/level — synchronize brightness/contrast adjustments across all viewports (Shift+W)
• Linked zoom — synchronize zoom levels across all viewports (Shift+Z)
• Cross-reference links — show slice position crosshairs across correlated viewports

10. Hanging Protocols

Hanging protocols automatically arrange series in the grid when a study is loaded. ImagiVue includes built-in protocols for common workflows:

• Mammography — laterality-aware layout (e.g., L-CC, R-CC, L-MLO, R-MLO)
• CT — axial, sagittal, coronal arrangement
• MR — sequence-based layout
• General radiography — optimized for single-image studies

Protocols are matched by modality, body part, and study description. When a matching protocol is found, the grid layout and series assignment are set automatically.

Managing Protocols

Open Settings (gear icon) and click Manage Hanging Protocols to:

• View all built-in and custom protocols with their priority levels
• Create new custom protocols with modality matching rules
• Edit or delete custom protocols
• Adjust protocol priority to control matching order

11. Cine Playback

For multi-frame series (cardiac, angiography, ultrasound), cine playback controls appear in the toolbar:

• Play / Pause — start or stop frame animation
• FPS slider — adjust playback speed from 1 to 60 frames per second
• Loop toggle — enable continuous looping
• Frame counter — shows current frame number and total frame count

12. Hierarchy Panel

The left panel shows loaded data organized by Patient, Study, and Series. Click to expand/collapse levels, and click a series to display it.

• Series show modality icons, descriptions, and instance counts
• Navigation arrows allow stepping through studies and series
• Hanging protocols automatically arrange related series when a study is selected
• The panel can be collapsed to a slim icon view with modality buttons for quick access
• Color DICOM images (ultrasound, photography) are displayed with full color

13. Settings

Click the gear icon in the top-right toolbar to access settings:

• AI Model — Select which segmentation model to use
• Auto-deselect tool — Automatically return to the Select/Pointer tool after placing an annotation
• GSDF Correction — Enable DICOM Part 14 Grayscale Standard Display Function for perceptually linearized grayscale rendering (Barten model). Recommended for diagnostic-quality display.
• Show DICOM warnings — Display informational messages about DICOM parsing
• Debug mode — Enable verbose logging and export mask/NIfTI data on each segmentation click (for development)
• Manage Hanging Protocols — View, create, edit, and delete hanging protocols

All settings persist across sessions via localStorage.

14. Research Platform

ImagiVue includes a comprehensive research platform for AI model evaluation and ground-truth labeling.

Project Management

• Create labeling projects with configurable label presets (custom names, colors, keyboard shortcuts)
• Organize annotation work by dataset and assign to team members
• Track completion progress per dataset and per annotator

Batch Labeling

• Efficient multi-image annotation with keyboard shortcuts and auto-advance
• Label presets bar for one-click annotation assignment
• Progress tracking with visual progress bar

Inference Evaluation

• Import predictions — load model output in NIfTI, COCO JSON, or raw JSON formats
• Comparison modes — overlay, difference map, or side-by-side view of ground truth vs. model predictions
• Difference maps — false positives (red), false negatives (blue), true positives (green)
• Live metrics — Dice coefficient, IoU, Hausdorff distance, sensitivity, specificity per case
• A/B model comparison — compare two models side by side with wins/losses/ties

Dataset Management

• Organize DICOM studies into labeled datasets
• Build evaluation cohorts by filtering on modality, body part, date range, and other metadata
• Client-side de-identification based on the DICOM PS3.15 Basic Profile
• Export datasets as JSON bundle, CSV, or markdown report with label distributions

Reporting

• Aggregate metrics across study sets with sortable per-case tables
• Dice histogram visualization
• Export reports as CSV or markdown
• Inter-rater agreement metrics (Dice/IoU) for comparing multiple annotators

Model Gallery

• Browse available AI models from vendors and the community
• Filter by modality and task type
• Import vendor model manifests
• Sponsored models are clearly labeled per FTC disclosure requirements

15. Demo Datasets

Sample DICOM datasets are available for testing. Access them through the file picker's demo button. Sources include:

• TCIA (The Cancer Imaging Archive) — lung CT, brain MR, mammography
• NIH Chest X-ray — frontal chest radiographs
• Synthetic/phantom — measurement accuracy testing

Datasets are fetched from public sources and loaded directly into your browser.

16. Privacy & Security

All patient data stays in your browser. ImagiVue uses a client-side-only architecture:

• DICOM files are parsed and rendered entirely in the browser
• AI models are downloaded once and run locally via WebGPU
• DICOMweb connections go directly from your browser to the hospital PACS — no data passes through ImagiVue's servers
• Annotations are stored in IndexedDB on your device
• No patient data is ever transmitted to cloud servers
• No analytics or tracking pixels touch patient data
• The cloud serves only application code (CDN) and AI model weights (CDN)

17. Troubleshooting