Improving medical diagnosis and treatment by using artificial intelligence to analyse medical images, such as x-rays, sounds like a great idea. But training Deep Neural Networks to do such work reliably requires the use of large amounts of personal medical data, and this brings a variety of challenges.
Dr. Vassili Kovalev and his team in the Department of Biomedical Image Analysis at the United Institute of Informatics Problems of the National Academy of Sciences of Belarus (UIIP NASB) explored two avenues to tackle the challenges: developing new methods and software for the automatic generation of artificial biomedical images that can be used for the training, and studying methods to create the kinds of malicious images used in ‘adversarial attacks’ and defences against them. For this the team needed very powerful computational resources, and gained access to them through BASNET, the Belarusian research and education network, the EU-funded EaPConnect project’s Enlighten Your Research (EYR) programme, and the pan-European GÉANT network. The EaPConnect team asked Dr. Kovalev about the work.
A story about this work is available on the R&E networking community’s In The Field blog site.
