Recent developments in artificial intelligence research have resulted in tremendous success in computer vision, natural language processing and medical imaging tasks, often reaching human or superhuman performance. In this thesis, I further developed artificial intelligence methods based on convolutional neural networks with a special focus on the automated analysis of brain magnetic resonance imaging scans (MRI). I showed that efficient artificial intelligence systems can be created using only minimal supervision, by reducing the quantity and quality of annotations used for training. I applied those methods to the automated assessment of the burden of enlarged perivascular spaces, brain structural changes that may be related to dementia, stroke, multiple sclerosis, and sleep.

Additional Metadata
Keywords neural networks, dementia, stroke, perivascular spaces, deep learning, machine learning, white matter hyperintensities
Promotor M. de Bruijne (Marleen) , M.W. Vernooij (Meike) , W.J. Niessen (Wiro)
Publisher Erasmus University Rotterdam
Sponsor This research was funded by The Netherlands Organisation for Health Research and Development (ZonMw) Project104003005.
ISBN 978-94-6375-865-9
Persistent URL hdl.handle.net/1765/126586
Organisation Department of Radiology
Citation
Dubost, F.P.G. (2020, May 8). Artificial Intelligence with Light Supervision: Application to Neuroimaging. Erasmus University Rotterdam. Retrieved from http://hdl.handle.net/1765/126586