There’s been a large increase in organizations using AI to extract insights from medical data. As the volume and diversity of medical imaging data increases, however, deep learning pipelines can easily be throttled by slow data loading. Getting DICOM (digital imaging and communications in medicine) images from a PACS into federated training applications like Clara can be a major workflow limiter.
We’ll describe the major components that impact the performance of a medical imaging AI workflow using Flywheel as the data curation and management platform and NVIDIA Clara Imaging framework for the software stack backed by Pure Storage FlashBlade with NVIDIA GPUs for the infrastructure.
You’ll see details on how Flywheel:
- Helps create automated workflows to take data from the acquisition device through clinical workflows or research analysis
- Eliminates the data bottlenecks that can arise from moving data between PACS, research systems, and AI clusters—whether working with tens, hundreds, or thousands of terabytes
- Enables efficient deployment of AI tools
- And more
Daniel Marcus, PhD, Chief Scientific Officer, Flywheel
Ravi Poddar, PhD, Director of EDA and HPC Solutions, Pure Storage