03. Core Screens

03. Core Screens

Routing is defined in src/App.tsx (React Router). After the first-run security wizard / login gate, the app exposes these routes:

Route	Screen	Purpose
/patients	Patient list	Create, search, edit, archive patients; preloaded demo patient
/patients/:patientId	Patient details	Clinical data + list of sessions
/patients/:patientId/sessions/:sessionId	Session view	Upload images (OD/OI), gallery, run analysis
/patients/:patientId/sessions/:sessionId/images/:imageId	Image analyzer	Multi-layer annotation canvas
/patients/:patientId/compare	Session comparison	Compare 2+ sessions of the same patient
/reports	Reports	Global list of generated PDF reports
/settings	Settings	Models, processing, appearance, security

The legacy Academy, Contribute and token-counter screens were removed — DIRD+ is a single-operator, fully local tool with no data-sharing or token economy.

Patient & session management

• Patients hold demographics and clinical context (diabetes type/duration, hypertension, dyslipidemia, medications).
• Sessions represent clinical visits and can be duplicated, edited, locked (after a final report), or combined for longitudinal analysis.
• A demo patient is preloaded so the system can be evaluated without real data.

Image analyzer (annotation canvas)

The canvas (src/components/canvas/, Konva) has 5 layers:

1. Original image
2. AI detections (bounding boxes)
3. AI segmentations (pixel masks)
4. Manual annotations
5. Measurements

Tools: selection, freehand, polygon, distance/area measurement, zoom, pan. Each layer has visibility / opacity / lock controls. Clinical overlays show retinal quadrants, the macular zone and the optic-disc area. The clinician can add, hide or modify detections on top of the AI output (corrections are flagged for traceability).

Settings

Sections under /settings:

• AI Models — download/activate ONNX detection models from Debaq/dird_models; add a custom ONNX model (file picker + model card, see 09. Model Interface); manage the local LLM assistant (download a small open-weight model, activate, test).
• Processing — confidence/sensitivity thresholds, CPU profile.
• Appearance — theme, primary color, language.
• Security — change passwords, view the encryption status. An encryption badge is always shown in the header.