Computer Vision / Video Analytics

From AI Research to Clinical Evaluation: NVIDIA Clara for Clinical Deployment at The Ohio State University

Deploying AI algorithms in a clinical environment is challenging as the tools and workflows of a data scientist’s AI lab are, in general, significantly different from those of radiologists in a clinical environment.

To develop clinically relevant AI algorithms, it is essential that researchers and engineers bring their expertise to integrate these algorithms to a radiologist’s clinical workflow. At the Department of Radiology at The Ohio State University’s Wexner Medical Center, researchers were able to achieve this workflow integration with NVIDIA Clara Deploy SDK.

The OSU team relied on a novel data augmentation technique and transfer learning  to first develop a state-of-the-art coronary artery classification algorithm for identifying any coronary artery showing signs of atherosclerosis.

Training images  were acquired from patients evaluated in the Emergency Rooms at the Wexner Medical Center for chest discomfort, one of the symptoms suggestive of coronary artery disease.

The annotated data set was then used to train a deep neural network, which achieved a high classification accuracy. The algorithm was subsequently considered for a feasibility study for integrating the AI algorithms in the routine clinical workflow.     

To deploy this solution in a clinical environment, the team selected an NVIDIA DGX Station and Clara Deploy SDK to seamlessly translate the AI algorithm into an application workflow that can be hosted in a hospital infrastructure where a radiologist can evaluate the results.


The key features of NVIDIA Clara Deploy SDK that made this possible were:  

  • DICOM Communication – NVIDIA Clara Deploy SDK includes a DICOM Adapter that allows receiving and sending of DICOM objects using standard DICOM protocols.
  • Crash proof Kubernetes- based deployment which is ideal for hospital environments and scalable for heterogeneous multi-modality workflows.  
  • Seamless integration into existing tools through the foundational framework. The data science team at OSU would need to only update the AI base container for the next radiology algorithm that they would like  to evaluate.

“The Clara platform has helped us deploy our AI algorithms and is the perfect solution to bridge the gap between AI research and the clinical environment,” said Dr. Vikash Gupta,  research scientist at The Ohio State University.

Read the full report below.

Read More >

Discuss (0)

Tags