Medical imaging annotation and reporting are critical for research, education, and the development of AI models in healthcare. However, current workflows are often:
- Manual and time-consuming: Annotating images and generating reports typically require significant expert effort.
- Fragmented: Tools for annotation, detection, and reporting are often separate, making workflows inefficient.
- Not easily accessible: Many solutions require specialized software or are not user-friendly for non-experts.
MedVision AI was created to address these issues by providing:
- An all-in-one, web-based platform for image annotation, detection, and AI-powered reporting.
- Instant switching between detection modes and persistent, shareable reports.
- A modern, intuitive interface for rapid experimentation and demonstration.
- Researchers and engineers working on medical imaging AI
- Educators and medical students / Medical Lab Radiologist learning about medical image analysis
- Developers building or testing annotation/reporting tools
Note: Many medical students and trainees now use large language models (LLMs) such as ChatGPT, Claude, or Gemini to help generate radiology reports or explanations. While these tools can be helpful for learning and brainstorming, they have important limitations (see below).
- Hallucinations: LLMs may generate plausible-sounding but factually incorrect or fabricated information.
- Lack of clinical validation: Reports generated by LLMs are not reviewed or validated by medical experts.
- No access to patient history: LLMs do not have access to real patient data, prior studies, or clinical context.
- Not a substitute for expert review: LLM-generated reports should never be used for clinical decision-making or patient care.
Always consult a qualified healthcare professional for medical advice and diagnosis.
A modern web app for medical image annotation, detection, and reporting using AI.
- Upload or select demo medical images (e.g., CT scans)
- Multiple detection modes: 2D bounding boxes, segmentation masks, points, 3D bounding boxes
- Medical imaging report generation (AI-powered)
- Persistent, collapsible report panel with copy/download
- Caching for fast mode switching and consistent results
- Modern, responsive UI with tooltips and loading indicators
- The app includes a demo image:
test_image.jpeg(CT scan) - This image is available as the first example image when the app loads
- You can also upload your own images (JPG, PNG, WEBP)
- Node.js (v16 or newer recommended)
- npm or yarn
-
Clone the repository:
git clone <your-repo-url> cd spatial-understanding-MV5
-
Install dependencies:
npm install # or yarn install -
Set up environment variables:
- Copy
.env.local.exampleto.env.local(if provided) - Add your Gemini API key or other required secrets to
.env.local
- Copy
-
Run the app:
npm run dev # or yarn devThe app will be available at
http://localhost:5173(or as shown in your terminal)
- Select a demo image from the left sidebar, or upload your own
- Choose a detection mode (2D bounding boxes, segmentation masks, etc.)
- View results instantly thanks to caching
- Generate a medical imaging report by clicking the Reporting button
- Copy or download the report from the right panel
- Built with React, TypeScript, and Tailwind CSS
- Uses Jotai for state management
- Easily extensible for new detection/reporting features
Pull requests and issues are welcome! Please open an issue to discuss major changes first.
Apache-2.0
Created by DR Arif Fahmi, MBBS, MSU Malaysia graduate
This application is for research and demonstration purposes only. It is not intended for clinical or diagnostic use. Do not use the outputs of this app to make medical decisions. Always consult a qualified healthcare professional for medical advice and diagnosis.