Clara Train SDK Documentation¶
Clara Train SDK is a domain optimized developer application framework that includes APIs for AI-Assisted Annotation, making any medical viewer AI capable and v4.0 enables a MONAI based training framework with pre-trained models to start AI development with techniques such as Transfer Learning, Federated Learning, and AutoML.
Using the open-source framework MONAI means Clara Train is now PyTorch-based compared to the previous versions of Clara Train before v4.0 which were based on TensorFlow. The concepts and Medical Model Archive (MMAR) for organization of model artifacts and Bring your own components (BYOC) continue to exist, with components from MONAI directly usable in Clara Train v4.0 MMARs.
If you are writing a paper and would like to reference Clara Train, the following example could be a way to do so:
NVIDIA Clara Imaging. https://developer.nvidia.com/clara-medical-imaging (2020).
- Clara Training Framework
- Pretrained Models
- AI-Assisted Annotation
- Federated learning
The Clara Train 4.0 release is based on NVIDIA’s container for PyTorch release 21.02 with support for NVIDIA Ampere GPUs. Here is a list of changes and additions in this version as well as links to key features:
The back end has been updated with MONAI, the Medical Open Network for AI and uses the PyTorch Ignite training loop. Please see Converting from Clara 3.1 to Clara 4.0 for details on converting artifacts from previous versions of Clara Train.
Federated learning now has homomorphic encryption tools to allow you to compute data while the data is still encrypted, and it still has easy server and client deployment through the use of an administration client like before although the back end is now implemented with NVFlare, which is usable for applications outside Clara as well.
Digital pathology in an example MMAR is now available including optimized data loading using cuCIM, which can tile large datasets on-demand and process them through a CUDA-enabled pipeline. It includes a trained fully convolutional classification network that works with whole-slide images. Along with other features in Clara, you can achieve up to a 10x speedup in training compared to other pathology pipelines.
For Jupyter Notebooks with detailed examples, see Notebooks for Clara Train SDK
You can use either MMARs to set up training configurations with json.
AIAA updates to Triton API
DeepGrow is introduced in AIAA to help with cold-start in any organs/objects of interest.