Doctors have used medical imaging for over a century to diagnose disease. Deep learning is giving this powerful tool a major upgrade.
This week, luminaries from the world of medical imaging will gather at MICCAI (the International Conference on Medical Image Computing and Computer-Assisted Intervention) in Granada, Spain, to present the latest research in these fields. It’s telling that 70 percent of the 400 papers to be featured at the conference use AI.
“AI is revolutionizing healthcare and recent reports show that medical image analysis is the top sector within this revolution,” said Alejandro Frangi, Diamond Jubilee Chair in Computational Medicine and director of CISTIB at the University of Leeds, U.K., and the General Chair for MICCAI 2018.
Now in its 21st year, MICCAI is widely recognized as the world’s leading research conference on medical imaging. This year’s event will be the largest yet, with a greater than 30 percent increase in both attendance and submitted research papers.
BraTS Boggles Brains
The work presented at MICCAI has never been more relevant — many countries worldwide are reporting a shortage in radiologists. Research seeks to automatically annotate medical images using AI, enabling doctors to diagnose disease more quickly. This technology has the potential to relieve the radiology resource gap, and help improve patient outcomes.
The BraTS (Multimodal Brain Tumor Segmentation) Challenge, the results of which were announced at MICCAI 2018, is one of the medical imaging research community’s most prestigious competitions. BraTS evaluates state-of-the-art methods for analyzing multimodal magnetic resonance imaging (MRI) scans of brain tumors.
From a field of almost 400 participants, NVIDIA Senior Research Scientist Andriy Myronenko won the challenge using autoencoder regularization, a deep learning segmentation technique.
Tackling the Deep Learning Data Shortage
NVIDIA’s applied research team presented a dozen papers at MICCAI 2018. They include work that addresses the shortage of accurate and reliable training data, a major roadblock to applying deep learning in medical research.
In basic terms, deep learning works by using large amounts of data to train machines to perform a specific task. The more data used to train the system, and the more diverse that data is, the better the deep learning system becomes at its task.
Using a DGX deep learning supercomputer, the team applied an AI technique called generative adversarial networks (GANs) to generate synthetic images, which can be used to train an AI-based medical imaging system. The deep-learning generated data can provide an automatable, low-cost source of diverse data that supplements real-world data.
And, because the images are synthetically generated, there are no patient data or privacy concerns. Medical institutions can easily share data they generate with other institutions, creating millions of different combinations that can be used to accelerate their work.
The team hopes their work can immediately help deep learning scientists generate new medical imaging data that can be used, in the end, to save lives.
Transforming Medical Imaging with Deep Learning
On the show floor and beyond, NVIDIA is infusing MICCAI 2018 with deep learning.
Hundreds of attendees participated in the two short hands-on courses run by the NVIDIA Deep Learning Institute. For those who missed out on the instructor-led training at MICCAI, there’s a wealth of DLI content targeting healthcare applications available online. Learn more and sign up at www.nvidia.co.uk/dli.
Among the highlights on the NVIDIA booth is NVIDIA Clara, an open platform that enables the medical industry to build and deploy breakthrough algorithms to create intelligent instruments and automate healthcare workflows.
At MICCAI, U.S.-based ultrasound technology specialist Cephasonics became the first to announce it’s adopting the NVIDIA Clara platform, including Clara AGX, to power its next-generation architecture for ultrasound AI and deep learning. NVIDIA Clara AGX is a revolutionary computing architecture based on the NVIDIA Xavier AI computing module and NVIDIA Turing GPUs.
Better Diagnosis, Better Outcomes
“Imaging is crucial for future precision medicine as it contributes to it rich, and arguably more specific and discriminative, biomarkers of health and disease,” said Frangi. “Moreover, with recent breakthroughs in deep learning, AI is enabling the holy grail of fully automated radiological data analysis.”