Holoscan Sample App Data for AI-based Bone Scoliosis Segmentation

Overview

This resource contains the spine ultrasound segmentation model for scoliosis visualization and measurement developed by Richard Brown at King's College London, as well as a sample ultrasound video.

Model

The AI model expects a gray-scale image of 256 x 256 and outputs a semantic segmentation of the same size with two channels representing

• bone contours with hyperechoic lines (foreground),
• hyperechoic acoustic shadow (background).

Note: The provided model is in ONNX format. It will automatically be converted into a TensorRT model (.engine) the first time it is processed by a Holoscan application.

Inputs

• INPUT__0 - Graysale ultrasound input image
  • shape=[3, 256, 256]
  • dtype=float32
  • range=[0,1]

Outputs

• OUTPUT__0 - 2-channel semantic segmentation of ultrasound image (hyperechoic lines, hyperechoic acoustic shadow)
  • shape=[2, 256, 256]
  • dtype=float32
  • range=[0,1]

Video Data

The sample data, originating from T. Ungi et al. (2020), is an ultrasound video with a spine identified by the model.

Note: the .avi file must be converted into a GXF tensor file using the convert_video_to_gxf_entities.py script on GitHub to be used with the VideoStreamReplayer Holoscan operator.

License

Refer to LICENSE.md supplied within, and to the license agreement for use of the sample data.