UM Data Science Research Seminar

Schedule 

12.00 - 12.30

Talk by Rianne Fijten (MAASTRO)

Title: Shared decision making and data science: perceived opposites that complement each other

Abstract: At first sight, shared decision making and data science seem like two vastly different topics. Shared Decision Making (SDM) is the process in which a patient and doctor decide together what treatment trajectory to choose for that individual patient. This is therefore a very qualitative approach, both in the clinic and in research. On the other hand, data science seems to be the opposite as it focusses on data quantitative statistics. Yet, this is far from the truth. In fact, SDM and data science can complement each other tremendously and can strengthen the clinical use-case for both. In this presentation I'll present the ways how these two seemingly opposite fields can complement each other and make each other stronger.

12:30-13:00

Talk by Zhenwei Shi (MAASTRO)

Title: Artificial intelligence big data in radiation oncology

Abstract: With the advances in AI technology, machine learning, especially deep learning, has achieved impressive progress recently especially in the domains such as face recognition and object detection. Although AI in general already forms the base of a pervasive suite of applications, deep learning applications in the medical imaging domain developed relatively slowly. One major challenge is that developing deep learning applications require large amounts of varied and high-quality training data. Actually, the healthcare domain as a whole does have “big data”. The problem is that most healthcare data are locked in local data repositories inside hospitals. Due to the concerns of political, ethical, legal, privacy and technical natures, these data are often not allowed to be shared among centres, especially among international centres where HIPAA and GDPR concerns need to be solved. These privacy concerns are important and worth to consider cautiously, but the negative aspect is that they have limited the healthcare domain from fully maximising the benefits of AI. To address the dispersal of big data in healthcare and leverage it for advance of AI in a privacy–preserving manner, federated/distributed machine learning have been proposed. The fundamental theory of federated learning is to exchange learning parameters (e.g., weights or gradients of deep neural networks) between locally accessed data, rather than exchanging patient-level data directly. The idea of federated learning could be described by a proverb – “bring the algorithm to the data, instead of the data to the algorithm”. In this presentation, we will talk about challenges and possible solutions of developing AI applications using big data in radiation oncology.

For questions or concerns, please contact us via info-ids@maastrichtuniversity.nl.

Maastro

Rianne Feijten

IDS

• Karen Zegers
• Kody Moodley
• Vikas Jaiman