Xilinx Launches Medical X-Ray on AWS
Launch of Classification Deep Learning Model and Healthcare AI Evaluation Board
Building training models for medical edge devices is possible.


In the medical field, demand for AI-based solutions that rapidly collect and analyze patient data is increasing. The COVID-19 pandemic, which has raised concerns about accumulated fatigue among medical staff, is accelerating this trend. Demand for telemedicine is also growing, requiring devices capable of AI processing at the edge.

On the 17th, Xilinx held an online press conference and unveiled, in collaboration with Spline.ai, a 'medical X-ray classification deep learning model' that is fully functional on Amazon Web Services (AWS) and the FPGA-based 'Xilinx Zynq® UltraScale+™ Healthcare AI Starter Kit.'

The deep learning model is built on the Zynq UltraScale+ MPSoC-based 'ZCU104 Evaluation Board' and utilizes Xilinx's Deep Learning Processor Unit (DPU), a powerful soft IP tensor accelerator that runs various neural networks such as disease classification and detection.

The Healthcare AI Starter Kit uses an open-source model that runs on the Python programming platform of the Zynq UltraScale+ MPSoC, allowing developers to tailor it to suit their diverse application needs.

Medical diagnostic and clinical equipment manufacturers and healthcare service providers can leverage the open source design and cloud expansion options above to deliver a wide range of clinical and radiology applications on mobile, portable, or point-of-care edge devices.Inning models can be developed and built quickly.

The deep learning model recently announced by Xilinx is trained with 'Amazon SageMaker' and deployed from the cloud to the edge using 'AWS IoT Greengrass', enabling remote updates of machine learning models, geographically distributed deployment of inference, and expansion across remote networks and wide areas.

“Xilinx and Spline.ai have developed a solution using Amazon SageMaker that enables accurate clinical diagnosis even with low-cost medical devices,” said Dirk Didascalou, Vice President of IoT at AWS. “Furthermore, by integrating AWS IoT Greengrass, they have further expanded the reach of telemedicine by enabling clinicians to easily upload X-ray images to the cloud without using physical medical devices.”

Currently, this solution is being used in detection systems for pneumonia, COVID-19, and other diseases due to its high accuracy and low inference latency. Xilinx and spline.ai trained a deep learning model using over 30,000 labeled pneumonia images and 500 COVID-19 images. This data is being used in public research at health and medical research institutions such as the National Institutes of Health (NIH), Stanford University, and MIT, as well as in hospitals and clinics around the world.

“FPGAs are suitable for low-power AI computation compared to general-purpose CPUs and GPUs, and it is easy to enhance security both in software and hardware,” said Subh Bhattacharya, leader of healthcare science and medical devices at Xilinx, who moderated the meeting. “Xilinx supports the development of adaptive, intelligent healthcare IoT solutions by providing semiconductors such as the Zynq UltraScale+ MPSoC as well as the AI inference development platform, Vitis.”