Nvidia announced this week at GTC Europe that Kings College London would deploy Nvidia’s DGX-2 and Clara medical imaging analysis platform to improve radiology and pathology practices. This is the first Clara deployment in Europe and evidence of AI’s growing adoption in health care where for years a wide variety of instruments, not least imaging devices, have been producing large datasets.
The hope now is to use AI – for training models and running inference engines – to make sense of the growing trove of medical data. “This is a huge opportunity to transform patient outcomes by applying the extraordinary capabilities of AI to ultimately make diagnoses earlier and more accurately than in the past,” said Sebastien Ourselin, head of the School of Biomedical Engineering and Imaging Sciences at KCL. “This partnership will combine our expertise in medical imaging and health records with NVIDIA’s technology to improve patient care across the U.K.”
Kimberly Powell, vice president of healthcare, wrote in a company blog, “NVIDIA and KCL will co-locate researchers and engineers with clinicians from major London hospitals that are part of the NHS Trust citywide network, including King’s College Hospital, Guy’s and St Thomas’, and South London and Maudsley. The trio of research, technology and clinicians will accelerate discovery of critical data strategies, targeted AI problems and speed deployment in the clinic.”
The DGX-2 system’s large memory and 2 petaflops of computing enable it to tackle the training of large, 3D datasets in minutes instead of days, according to Nvidia. The Clara platform bundles a variety of AI and medical imaging analysis capabilities.
At Clara’s launch in the spring, Nvidia CEO Jensen Huang, said “This technology is used everywhere, CT, MRI, PET scans. We can now reconstruct images better than ever before. We can visualize images in a way that release more insight with cinematic rendering. The entire software stack, the solvers and libraries, integrated into [Clara],” said Huang.
“The unfortunate thing is there around 3 million medical instruments installed, but only 100k sold each year. It would take 30 years to update everything,” said Huang who emphasized that Clara was, among other things, architected to work with older legacy datasets making it possible to “virtually update every system that’s out there.” At GTC, Huang showed 15-year-old ultrasounds images that had been re-rendered in color and displaying features with far more realism and accuracy. There was also footage of heart function in which the anatomy and output were better identified and quantified.
As HPCwire noted in coverage of the Clara launch, AI-driven analysis of medical images has certainly been revolutionizing the insight the medical community can extract from medical images. Making such processing available to the installed base of instruments could dramatically improve their usefulness. Most of the imaging equipment in hospital is already connected to a network.
Powell wrote in her blog yesterday, “[T]raining at scale is tricky, but the DGX-2 can enhance medical imaging AI tools like Niftynet, a TensorFlow-based open-source convolutional neural network platform for research in medical image analysis and image-guided therapy developed at KCL.
“Working with KCL’s clinical network to crack the technical and data governance issues of federated learning could lead to breakthroughs such as more precisely classifying stroke and neurological impairments to recommend the best treatment or automatic biomarker determination.
“From development to deployment, NVIDIA and KCL plan to streamline AI while building the necessary tools, infrastructure and best practices to empower the entire clinical ecosystem.”