News | Artificial Intelligence | October 23, 2019

NVIDIA and King's College London Debut First Privacy-preserving Federated Learning System

Federated learning is a type of deep learning or artificial intelligence that distributes training data across multiple locations

NVIDIA and King's College London Debut First Privacy-preserving Federated Learning System

Image courtesy of NVIDIA

October 23, 2019 — To help advance medical research while preserving data privacy and improving patient outcomes for brain tumor identification, NVIDIA researchers in collaboration with King’s College London researchers recently announced the introduction of the first privacy-preserving federated learning system for medical image analysis. NVIDIA is working with King’s College London and French startup Owkin to enable federated learning for the newly established London Medical Imaging and AI Centre for Value Based Healthcare.

This new paper was presented at the International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI), one of the world’s top conferences on medical imaging, Oct. 13-17 in Shenzhen, China. In the work, NVIDIA and King’s College London researchers describe in detail how they developed their technique.

Federated learning is a learning paradigm that allows developers and organizations to train a centralized deep neural network (DNN) with training data distributed across multiple locations. This makes it possible for organizations to collaborate on a shared model, without needing to directly share any clinical data. 

“Federated learning allows collaborative and decentralized training of neural networks without sharing the patient data,” the researchers stated in their paper. “Each node trains its own local model and, periodically, submits it to a parameter server. The server accumulates and aggregates the individual contributions to yield a global model, which is then shared with all nodes.”

Although federated learning can provide high security in terms of privacy, there are still ways to reconstruct data by model inversion, the researchers explained. To help make federated learning even safer, the researchers investigate the feasibility of using the ε-differential privacy framework, a way to formally define privacy loss, to protect patient and institutional data with a strong privacy guarantee. To ensure patient privacy is priority, differential privacy and other state-of-the-art privacy protection techniques are being built into the Owkin architecture.

The experiments for this breakthrough were performed on brain tumor segmentation data from the BraTS 2018 dataset. The BraTS 2018 dataset contains magnetic resonance imaging (MRI) scans of 285 patients with brain tumors.

The dataset is used here for evaluating federated learning algorithms on the multi-modal and multi-class segmentation task. On the client side, the team adapted a state-of-the-art training pipeline originally designed for data-centralized training and implemented it as part of the NVIDIA Clara Train SDK.

For training and inference, the team used NVIDIA V100 Tensor Core GPUs.

When comparing federated learning to a data-centralized system, the proposed approach can achieve a comparable segmentation performance without sharing institutional data. 

Moreover, the experimental results show a natural tradeoff between privacy protection and the quality of the trained model. Still, with the sparse vector technique, the federated learning system can provide rigorous privacy protection with only a reasonably small cost in model performances.

Deep learning is a technique for automatically extracting knowledge from medical data, the NVIDIA team explained. Federated learning has the potential of effectively aggregating knowledge across institutions learned locally from private data, thus further improving the accuracy, robustness, and generalization ability of the deep models, they added.

“This research is an important step towards the deployment of secure federated learning, which will enable data-driven precision medicine at large scale,” the NVIDIA researchers stated.

For more information: www.nvidia.com

Related Content

Image courtesy of GE Healthcare

Feature | Mobile C-Arms | July 08, 2020 | By Melinda Taschetta-Millane
Moblie C-arms have seen several advances over the past de
At the American Association of Physicists in Medicine (AAPM) 2019 meeting, new artificial intelligence (AI) software to assist with radiotherapy treatment planning systems was highlighted. The goal of the AI-based systems is to save staff time, while still allowing clinicians to do the final patient review. 
Feature | Treatment Planning | July 08, 2020 | By Melinda Taschetta-Millane
At the American Association of Physicists in Medicine (AAPM) 201
 Many patients with severe coronavirus disease 2019 (COVID-19) remain unresponsive after surviving critical illness. Investigators led by a team at Massachusetts General Hospital (MGH) now describe a patient with severe COVID-19 who, despite prolonged unresponsiveness and structural brain abnormalities, demonstrated functionally intact brain connections and weeks later he recovered the ability to follow commands

Getty Images

News | Coronavirus (COVID-19) | July 08, 2020
July 8, 2020 — Many patients with severe coronavirus disease 2019 (...
Several drivers will contribute to the growth of the teleradiology market in terms of penetration, revenue and read volumes over the next five years

Getty Images

Feature | Teleradiology | July 08, 2020 | By Arun Gill
Last year was a record year for the global...
This is Figure 2 from the article in Radiology: Acute encephalopathy. A 60 year-old-man without history of seizures presenting with convulsion. (A-B) Multifocal areas of FLAIR hyperintensity in the right cerebellum (arrows in A), left anterior cingular cortex and superior frontal gyrus (arrows in B). (C-D) Restricted diffusion in the left anterior cingulate cortex, superior frontal and middle temporal gyrus (arrows in D) and right cerebellum (arrows in E), consistent with cerebellar diaschisis. F)  #COVID19

This is Figure 2 from the article in Radiology: Acute encephalopathy. A 60 year-old-man without history of seizures presenting with convulsion. (A-B) Multifocal areas of FLAIR hyperintensity in the right cerebellum (arrows in A), left anterior cingular cortex and superior frontal gyrus (arrows in B). (C-D) Restricted diffusion in the left anterior cingulate cortex, superior frontal and middle temporal gyrus (arrows in D) and right cerebellum (arrows in E), consistent with cerebellar diaschisis. F) No hemosiderin deposits in gradient echo sequences.

Feature | Coronavirus (COVID-19) | July 06, 2020 | Dave Fornell, Editor
Four recent radiology studies, from New York, Italy, Iran and China, show how...
A patient implanted with the Axonics System can undergo MRI examinations safely with radio frequency (RF) Transmit Body or Head Coil under the conditions outlined in the Axonics MRI Conditional Guidelines.

A patient implanted with the Axonics System can undergo MRI examinations safely with radio frequency (RF) Transmit Body or Head Coil under the conditions outlined in the Axonics MRI Conditional Guidelines.

News | Magnetic Resonance Imaging (MRI) | July 02, 2020
July 2, 2020 — Axonics Modulation Technologies, Inc., a medical technology company that has developed and is commerci
This data represents wave 2 of a QuickPoLL survey conducted in partnership with an imagePRO panel created by The MarkeTech Group (TMTG), regarding the effects of COVID-19 on their business

Getty Images

Feature | Coronavirus (COVID-19) | July 01, 2020 | By Melinda Taschetta-Millane
A 3-D ultrasound system provides an effective, noninvasive way to estimate blood flow that retains its accuracy across different equipment, operators and facilities, according to a study published in the journal Radiology.

Volume flow as a function of color flow gain (at a single testing site). For each row the color flow c-plane and the computed volume flow are shown as a function of color flow gain. The c-plane is shown for four representative gain levels, whereas the computed volume flow is shown for 12–17 steps across the available gain settings. Flow was computed with (solid circles on the graphs) and without (hollow circles on the graphs) partial volume correction. Partial volume correction accounts for pixels that are only partially inside the lumen. Therefore, high gain (ie, blooming) does not result in overestimation of flow. Systems 1 and 2 converge to true flow after the lumen is filled with color pixel. System 3 is nearly constant regarding gain and underestimates the flow by approximately 17%. Shown are mean flow estimated from 20 volumes, and the error bars show standard deviation. Image courtesy of the journal Radiology

News | Ultrasound Imaging | July 01, 2020
July 1, 2020 — A 3-D ultrasound