MIDL Vision 2020

We test whether CNNs could be trained on explicit textural representations of the images in a fairly complex neuroimage segmentation task. We use the well-established local binary patterns (LBP) algorithm which mainly captures local textures. While there might be some shape information left after processing, this would be minimal. E.g. (a) is the original MRI scan, (b) and (c) are textural representations calculated using the LBP algorithm, and (d) are example segmentation labels we wish to predict. 28 Poster Presentation