The healthcare sector has been on a journey to use advanced analytics to improve all aspects of their business for quite some time. This includes everything from operations, diagnostics, and research to risk assessments. Artificial Intelligence (AI), and more specifically deep learning, is one of those advanced analytics tools. Radiologists, for example, are benefiting from the advances in deep learning algorithms, with most of the data they collect already in a digital format. The digital composition makes it easier to be processed with AI algorithms, thus turning medical centers and hospitals into places where computer vision technology becomes part of how radiologists augment their work.

Another promising area AI and deep learning are being used is in assisted diagnostics. Radiology images are huge, complex, and include things like CT scans, PET scans, and MRIs – making viewing the images and providing an assessment time intensive for the radiologist. But in order to train a reliable model, you need a lot of images.

The images also must be preprocessed before they can be fed to an AI model. This means that the underlying infrastructure must be able to accommodate the data pipeline as well as deliver the processing speed needed to perform analytics in a reasonable amount of time.

Cisco and SAS solution

Cisco and SAS collaborated on a reference architecture that addresses both the data pipeline and the infrastructure requirements for delivering timely video analytics results.

The SAS Viya platform and Cisco solution supports computer vision modeling from start to finish, using SAS Visual Data Mining and Machine Learning to prepare the image data set and train the deep learning model. Cisco provides the accelerated compute infrastructure, with the Cisco UCS C480 ML server with 8 NVIDIA V100 GPUs, to deliver the performance needed to rapidly train, deploy and score the deep learning models.

The SAS Platform architecture for deep learning uses massively parallel processing and parallel symmetric multiple processors (SMPs) with multiple threading for extremely fast processing.

Because of the heavy parallel processing required by the deep learning algorithms, the solution also benefits greatly from the GPU acceleration provided by the UCS C480 ML, delivering a 90% reduction in training time compared to using no GPU.

Cisco and SAS platform architecture
SAS platform architecture for deep learning on Cisco UCS [click image to enlarge]
The SAS platform architecture for deep learning on Cisco UCS delivers:

  • An efficient and scalable video and image analytics platform
  • Support for a variety of video and image analytics solutions
  • Extensive storage capacity and the accelerated compute needed for large scale training

The Cisco and SAS solution provides a tested and validated combination of hardware and software to create and deploy biomedical image analytics. The solution lets you quickly deploy your biomedical image analysis infrastructure with a purpose-built server for deep learning. The pretrained SAS deep learning models simplify the training steps. The solution provides training workload at scale for very large data sets.

Next step

Read the SAS Visual Data Mining and Machine Learning on Cisco UCS solution overview

Visit Cisco big data analytics solution page

Visit Cisco AI solution page

Connect on Twitter @FrancoiseBRees