Xilinx, Spline.AI and Amazon team up on ML x-ray classification

October 14, 2020 //By Nick Flaherty
Xilinx teams for cloud-based open source Covid-19 X-ray classification
Xilinx has teamed with Spline.AI and Amazon on a cloud-based adaptive machine learning model that can recognise different lung infections from X-rays

Xilinx has launched a fully functional medical X-ray classification deep-learning open source model and a reference design kit running on its FPGAs on Amazon Web Services (AWS). This can allow medical equipment makers to help improve diagnostics of a range of repiratory infections from pneumonia to Covid-19

The model was developed with Spline.AI and Amazon Web Services (AWS) GreenGrass IoT service. The model is deployed on the Xilinx Zynq UltraScale+ MPSoC device based ZCU10. It makes use of the Xilinx deep learning processor unit (DPU), a soft-IP tensor accelerator, to run a variety of neural networks, including classification and detection of diseases.

“We were working closely with Spline.ai before Covid hit when were working on an X-ray classification model for pneumonia and then we started on a reference design kit and then we started looking at potential data for Covid and an open source model on Github,” said Subhankar Bhattacharya, Lead for Healthcare & Sciences at Xilinx.

“We managed to acquire a curated dataset from a number of sources to distinguish between pneumonia and other diseases and then we added Covid-19,” he said. “That was extended to include AWS Greengrass for scalability and the reference design kit allows medical equipment makers to create other platforms with different radiological models.”

The data came from public research by healthcare and research institutes such as National Institute of Health (NIH), Stanford University, and MIT, as well as other hospitals and clinics around the world.

“Right now it supports AWS GreenGrass but it could be in other cloud providers or purely and edge solution. It could also be complied entirely under Vitis and fed in the flow and retrain as more data comes along,” he said.

"This uses the VC104 as the edge device and the model has been trained with 30,000 images for pneumonia and 500 for Covid-19 and we plan to update that every


Vous êtes certain ?

Si vous désactivez les cookies, vous ne pouvez plus naviguer sur le site.

Vous allez être rediriger vers Google.