Essential Citronnier
← Back
Open Source DICOM ViewerMIT LicenseSwiftUI · Metal · Gemma 4 · MLX

AI-Powered DICOM Annotation on Your Mac.

Fully local. Fully private. Powered by Gemma 4 E4B running on Apple Silicon via MLX. Analyze images, detect structures, describe ROIs — no cloud, no uploads.

Download for macOSView on GitHub

macOS 14.0+ (Sonoma) · Apple Silicon · 16GB+ RAM recommended

ODV-Annotate multi-panel DICOM viewer
ODV-Annotate AI annotation with Gemma 4

DICOM Viewer Meets On-Device AI

Everything you need in a medical image viewer, plus AI annotation that runs entirely on your machine.

01Native Performance

SwiftUI + Metal Native App

Native Performance with SwiftUI & Metal

Built with SwiftUI and Metal. No Electron, no web views — just native macOS speed with GPU-accelerated rendering. Custom pure-Swift DICOM parser ensures the first image appears before the rest of the study finishes loading.

  • SwiftUI-based native macOS interface
  • Metal GPU-accelerated rendering
  • Custom pure-Swift DICOM parser for fast loading
  • Background loading with instant first image display
02Gemma 4 AI

On-Device AI Annotation

On-Device AI Annotation with Gemma 4

Google's Gemma 4 vision-language model runs locally on Apple Silicon GPU via MLX. Analyze images, detect structures, describe ROIs, identify abnormalities — no internet required, no cloud uploads, 100% private.

  • Gemma 4 E4B 4-bit quantized model running locally
  • Full image analysis with anatomical structure detection
  • Natural language ROI descriptions on selection
  • Abnormality detection with severity and confidence scores

AI Analysis Modes

Four modes powered by Gemma 4 for different workflows.

Clinical

Only clinically significant findings. Maximum 3 bounding boxes for quick assessment.

Detailed Anatomy

All visible anatomical structures labeled and boxed. Minimum 3 structures for comprehensive mapping.

Abnormality Only

Pathological findings only. Up to 2 abnormalities with severity and confidence.

Educational

Labeled anatomy for learning. 3-5 structures with detailed descriptions.

No internet requiredNo cloud uploadsNo data collectionAll inference on Apple Silicon GPU

Get Started

Up and running in seconds.

Download for macOS

  • 1.Download the latest release, open the DMG, and drag to Applications.
  • 2.Signed and notarized — no Gatekeeper warnings.
  • 3.AI model downloads automatically on first use (~4.5GB).
Download for macOS

Build from Source (optional)

# Clone and build

git clone https://github.com/Essential-Citronnier/ODV-Annotate.git

cd ODV-Annotate

./scripts/package_app.sh

Built by Researchers

Built on Prof. JoonNyung Heo's OpenDicomViewer open source. MIT licensed open-source project.

JoonNyung Heo, MD, PhD

Original OpenDicomViewer Author

Assistant Professor of Neurology, Yonsei University College of Medicine. Stroke neurologist & neurointerventionist. MIT Technology Review Innovators Under 35 (Korea, 2021).

Yonsei Stroke TeamMIT Innovators Under 35Neurointervention

Daeseong Kim, MD

AI Annotation Developer

Research Scientist, Computational & Translational Oncology. Yonsei University College of Medicine (MD, 2025, Summa Cum Laude). First-author in Clin Mol Hepatol. Built the Gemma 4 E4B on-device annotation layer.

Yonsei MedicineDean's AwardTranslational OncologyFounder of Essential Citronnier
ODV-Annotate

Get Started Now

OpenDicomViewer-Annotate is open source under the MIT License. Built with SwiftUI, Metal, and Gemma 4.

Download for macOSGitHub