On Nov 26th 2020 we present the last two lectures of our series.
Machine Learning for the Detection of Brain Abnormalities
by Dr. Hugo Kuijf (Image Sciences Institute, UMC Utrecht)
Magnetic Resonance Imaging as a clinical diagnostic method is one of the greatest innovations of the twentieth century. In Europe, millions of images are made yearly and the number is steadily increasing. Assessment of medical images relies on visual inspection, which can be time-consuming and subjective. Automated machine learning solutions have proven essential for reliable detection and quantification of brain pathology.
Brain abnormalities–associated with stroke, dementia, and aging–have been a key application for machine learning solutions in medical image analysis. Various automated analysis techniques have been developed, to provide quantitative measurements and replace time-consuming, observer-dependent delineation procedures. Such techniques exist for white matter hyperintensities (WMH), microbleeds, microinfarcts, intracranial aneurysms, and more. They are based on classic machine learning techniques, including k-nearest neighbours, support vector machines, or random forest classifiers, but in recent years (deep) neural network techniques have been introduced using convolutional neural networks. Objectively assessing the performance of these techniques can be done within scientific contests/challenges, for example the WMH Segmentation Challenge (https://wmh.isi.uu.nl/), the MR Brain Segmentation Challenge (https://mrbrains18.isi.uu.nl/), or the Aneurysm Detection And segMentation challenge (http://adam.isi.uu.nl/).
Deep Learning to Boost Medical Systems’ Performance
by Dr. Nicola Pezzotti (Philips Research and TU Eindhoven, Eindhoven)
Machine Learning, and Deep Learning in particular, is beyond recent advancements in medical image analysis.
From segmentation tasks, to tumor grading and workflow support, deep learning methodologies are providing superior performance and are becoming more common in clinical research and practice.
While these techniques are mainly used as post processing tool, a new and exciting trend is to apply deep learning techniques early in the medical image formation chain. In this context, new machine learning algorithms are applied to boost the performance of medical systems, either by providing better image quality, faster acquisition or lower radiation dose.
In this talk, I will present the state of the art in the domain, highlighting some of the most exciting research directions and highlighting the gaps in literature that limit the application of deep learning in medical systems. Moreover, I will present the success story of the “Philips & LUMC” team in the fastMRI challenge, a competition organized by NYU Langone Health and Facebook AI Research to reduce the acquisition time of MRI scanners.
Share this Post