MedShapeNet

We are looking for more shapes for MedShapeNet. If you want to contribute medical shapes, please get in touch with us — as a reward for your efforts, we will include you as a co-author on our next MedShapeNet paper.

Prior to the deep learning era, the wide application of statistical shape models in the medical domain was evidence that shapes are common in medical imaging. Inspired by the well-known ShapeNet and Princeton ModelNet in computer graphics and computer vision, the goal of MedShapeNet is to promote the translation of methods developed in that context to clinical applications. MedShapeNet contains over 100,000 medical shapes — bones, organs, vessels and muscles among them, as well as surgical instruments.

The shapes are derived directly from real-world data, come paired with annotations and span 23 datasets. They are reachable through the dataset browser and a Python API, and serve image processing tasks as much as applications in virtual, augmented and mixed reality and in 3D printing. The collection was announced on LinkedIn.

Watch the introduction on YouTube — a German-language film on MedShapeNet produced by the Centre for Virtual and Extended Reality in Medicine (ZvRM).

Downloading the shapes

Individual shapes are browsed and downloaded in the MedShapeNet browser. If you run into problems with the Streamlit site, please get in touch with Gijs Luijten.

For the whole collection, save the download index — one direct download URL per shape, 106,101 of them — and hand it to wget:

wget --content-disposition --trust-server-names -i MedShapeNetDataset.txt