Features

The DRAW project aims to provide general purpose automatic segmentation for radiotherapy. Thus key features are:

• Models are trained using a robust nn-UNET-based convolutional neural network. This architecture has demonstrated its accuracy in multiple segmentation tasks.
• All models are generated using curated datasets from patients after consent. De-identified data is used, and the de-identified data is available for download at https://chavi.ai.
• Model performance benchmarks are available publicly for review.
• Detailed delineation guidelines followed for the structures in each model are provided to the user.
• The system has a client-server architecture, so there is no need to install an expensive computer at the premises to use the system.

Client Features

The client installed on the local computer performs several functions:

• Allows users to create automatic segmentation templates from models available in the DRAW catalog.
• Allows users to pre-specify rules which will automatically associate the templates with the imaging data.
• Ensures that users can manually associate templates with a case where greater customization is needed.
• Automates processing of the DICOM data from your imaging system.
• De-identifies DICOM data before sending it to the server, ensuring patient privacy.
• Re-identifies DICOM data received by the server.