Deep 3D Convolutional Encoder Networks with Shortcuts for Multiscale Feature Integration Applied to Multiple Sclerosis Lesion Segmentation.Our model is a neural network that consists of two interconnected pathways, a convolutional pathway, which learns increasingly more abstract

higher-level image features, and a deconvolutional pathway, which predicts the final segmentation at the voxel level.We have evaluated our method on two publicly available data sets (MICCAI 2008 and ISBI 2015 challenges) with the results showing that our method performs comparably to the top-ranked state-of-the-art methods, even when only relatively small data sets are available for training.In addition, we have compared our method with five freely available and widely used MS lesion segmentation methods (EMS, LST-LPA, LST-LGA, Lesion-TOADS, and SLS) on a large data set from an MS clinical trial. The results show that our method consistently outperforms these other methods across a wide range of lesion sizes.

The automatic segmentation of MS lesions is a very challenging task due to the large variability in lesion size, shape, intensity, and location, as well as the large variability of imaging contrasts produced by different scanners used in multi-center studies. However, outliers are often not specific to lesions and can also be caused by intensity inhomogeneities, partial volume, imaging artifacts, and small anatomical structures such as blood vessels, which leads to the generation of false positives. To overcome those challenges, supervised methods require large data sets that span the variability in lesion appearance and careful pre-processing to match the imaging contrast of new images with those of the training set. Library-based approaches have shown great promise for the segmentation of MS lesions, but do not scale well to very large data sets due to the large amount of memory required to store comprehensive sample libraries and the time required to scan such libraries for matching patches.