Open-source · nnU-Net v2 · Client-Server Architecture
What is DRAW?
It is built on the nnU-Net v2 framework and uses a distributed client-server architecture with a decoupled inference layer, enabling on-demand use of GPU resources. DRAW was designed specifically for deployment in hospitals in low- and middle-income countries where commercial autosegmentation systems are often prohibitively expensive and difficult to customise.
The Challenge
Automatic segmentation saves valuable time and reduces variability in contouring. However, the key challenges in adopting commercial automatic segmentation systems include high cost, vendor lock-in, a lack of representation of the local population in the training dataset, and limited capability to customise models.
In low- and middle-income countries with limited resources and technical know-how, deploying these models is not trivial. Most centres lack the technical know-how and resources to deploy inference pipelines locally, even after purchasing adequate computing hardware.
How It Works
The DRAW Client, deployed on-site using Docker, automatically processes incoming DICOM images, de-identifies them using HIPAA Safe Harbor standards, and uploads them to the API Server. The server orchestrates ephemeral GPU containers that run nnU-Net v2 inference. Once segmentation is complete, the RT-Struct is returned to the client, which re-identifies it and sends it to central storage.
The inference layer is decoupled from the server, so expensive GPU resources are used only on demand — suiting radiotherapy workloads where images are acquired over daytime working hours.
Why DRAW?
Open Source
No licensing costs, no vendor lock-in. DRAW is built entirely on open-source software using the nnU-Net v2 framework. The full codebase is available on GitHub. The system is decoupled from commercial treatment planning systems, so the output can be used irrespective of the vendor — Varian, Accuray, Elekta, and others.
No Local GPU
Inference runs on ephemeral GPU containers managed by the cloud server. Hospitals do not need to purchase, maintain, or house expensive, energy-intensive GPU hardware on-site. The client only needs a standard desktop workstation with Docker and a network connection. GPU resources are used only on demand, suiting radiotherapy workloads where images are acquired during daytime working hours.
Customisable
Hospitals can train and deploy custom models on their own patient data. DRAW supports site-specific configurations and structure definitions following the TG-263 standard. Contouring practices vary across centres — DRAW accommodates multiple ways of delineating the same volume. New models can be trained within one week if a curated dataset is available.
Transparent Metrics
Every model in the DRAW catalog publishes its training data size, validation dataset size, and per-structure spatial overlap metrics including Dice similarity coefficient (DSC), Dice Jaccard coefficient (DJC), 95th percentile Hausdorff distance (HD95), surface DSC, and mean distance to conformity (MDC). No black boxes.
Featured Models
View All →Register Your Interest
Interested in deploying DRAW at your institution? Register your interest and our team will get in touch.
Register →Explore Models
Explore the full catalog of 60 segmentation models covering 142 structures.
Browse Catalog →See the Impact
See how DRAW is performing in real clinical deployments across multiple centres.
View Impact →