Discover privacy-preserving AI in healthcare

Overview of techniques for secure, federated, and privacy-preserving machine learning with a focus on healthcare and medical imaging.

Abstract:

The webinar and live discussion will provide an overview of techniques for secure, federated and privacy-preserving machine learning with a focus on healthcare and medical imaging.

The widespread clinical deployment of artificial intelligence algorithms will require large-scale patient datasets for training and validation. However, patient privacy concerns, alongside legal and ethical requirements mandate the protection of personally identifiable health data. At the same time, algorithm creators and owners wish to protect their models against unwarranted exploitation or theft.

To reconcile the unmet needs for privacy and asset protection while enabling training of clinically useful machine learning models on large datasets, secure and privacy preserving machine learning techniques are being developed.

The webinar will introduce viewers to these methods including federated learning, differential privacy, homomorphic encryption, secure multi-party computation, on-device privacy and distributed computation approaches with cryptographic security guarantees, and provide insight into their technical challenges and limitations.

The webinar is suited for a broad audience including healthcare professionals, machine learning researchers and practitioners and researchers of associated disciplines like data ethics or systems security. A prior understanding of machine learning is desirable, but no knowledge of cryptography is required. The speakers are experts in the field of healthcare and radiology, machine learning, secure and private artificial intelligence and blockchain systems. Viewers will have the opportunity to ask questions beforehand and during the open panel discussion at the end of the webinar.

This webinar will cover some of the work published in Nature Machine Intelligence Journal :

Secure, privacy-preserving and federated machine learning in medical imaging. Read the paper here: https://www.nature.com/articles/s42256-020-0186-1

Presenters: Prof Dr Rickmer Braren, Dr Georgios Kaissis, MHBA, and Dr Jonathan Passerat-Palmbach

Dr Braren is an adjunct professor of radiology and senior attending radiologist at the Technical University of Munich, Germany. He specializes in abdominal oncologic imaging with a focus on pancreatic cancer imaging. His research focus lies on the development and clinical translation of novel imaging methods and of medical image analysis algorithms and on the integration of multiparametric clinical and imaging data.

Dr Kaissis is a specialist diagnostic radiologist at the Technical University of Munich, machine learning researcher at Imperial College London and a research scientist at OpenMined. His research focuses on probabilistic methods and differentiable programming as well as on the design and deployment of robust, secure, fair and transparent machine learning algorithms to medical imaging workflows using next-generation privacy-preserving machine learning methods.

Dr Jonathan Passerat-Palmbach: more info to come