Ultrasound Ovarian Segmentation Sample Application#

GitHub Readme

Note

This application is for reference and evaluation purposes only. It is not intended for direct use in clinical or diagnostic environments and is not validated for such a purpose.

The application demonstrates how the OpenVINO™ Toolkit (OpenVINO™) uses Intel hardware acceleration (Intel iGPU, with CPU fallback) for AI-based real-time ovarian tumor segmentation in 2D ultrasound. It offers two interchangeable models, selected with -Arch: a default promptless SegFormer-B5 (DS2Net) and an optional prompted SAM-256 pipeline (YOLO bounding box → SAM encoder → mask decoder). An architecture that decouples capture, inference, and display keeps the displayed video smooth regardless of inference speed, and a built-in GPU governor enforces the KPI of keeping Intel iGPU utilization below 80%.

Supported inputs:

  • Video file (for demonstrations and benchmarking)

  • USB camera or webcam

  • Folder with ultrasound still images (paced like a slideshow)

Next Steps#

  • See Model Preparation to produce the OpenVINO™ Intermediate Representation (IR) that the application loads. The application does not include a trained model.

  • See Get Started for step-by-step setup and run instructions, including the web dashboard (UI), which is the primary way to control the application.

  • See Runtime Configuration for all flags and environment variables that the application exposes.

  • See Troubleshooting for common issues.

  • See Release Notes for version history.