Machine Learning in Brain Data Processing

Collaborations

2016-Present. Univ. of Bergen, Norway (Arvid Lundervold)

Machine learning, particularly deep neural networks and convolutional neural networks, is behind an ongoing revolution in both clinical and preclinical imaging. Although deep learning has been studied since the early 1980s, the methods did not have great impact until about 2011, when techniques enabling effective learning and layerwise pretraining, usage of fast general purpose graphics processors, and large labelled datasets all came together. Over the past few years the field has revolutionized computer vision, especially since 2015, when deep learning achieved ’superhuman’ levels of perception. In medical fields, deep learning techniques have recently been shown to outperform traditional image analysis approaches in a variety of applications. Deep learning was named one of“10 breakthrough technologies” by MIT Technology Review in 2013 and a Method to watch by Nature in 2016. Applied to medical imaging it may become one of the most disruptive technology radiology has seen since the advent of digital imaging. Successful deployment of these powerful techniques requires methodological research, well-formulated and important research questions from relevant domains, and access to good and big datasets. This is the context in which our project aims to make significant imaging-based contributions to several fields within medicine: applications such as brain tumours, neurodegeneration, multiple sclerosis and stroke, diseases of kidney function and structure, endometrial and prostate cancer, and translational studies related to live cell imaging and histopathology, and to methodological research.

The main goal of the project is to apply, to research, and deploy next generation medical image and data analysis based on machine learning, particularly deep learning and convolutional neural networks, for fast and accurate image segmentation, multimodal image registration, motion correction and probabilistic predic- tion. The ultimate aim is to improve diagnosis, prognosis and therapy in diseases where imaging is an important source of information.

More infromation on here.

Publications

. Assessing functional brain connectivity changes in cognitive aging using RS-fMRI and graph theory. In Proc. of 21st International Symposium on Mathematical Programming, Berlin, August 19-24, 2012.

PDF Project bibtex