Longitudinal Multiple Sclerosis Lesion Segmentation
using Multi-View Convolutional Neural Networks
In this work we implemented a fully automatic system for the challenging task of MS lesion segmentation in brain MRI.
Our system can help diagnosis and patient follow-up, while reducing time consuming manual labor from the expert physician.
The system is composed of multiple pre-processing steps and a voxel-level prediction step using Convolutional Neural Networks (CNNs).
Our CNNs process longitudinal data, a novel contribution in the domain of MS lesion analysis.
In addition, our CNNs make use of multiple contrast images and multiple orthogonal views from volumetric data.
Our system was evaluated on the 2015 ISBI MS Lesion Segmentation
Challenge data set, in which we obtained state-of-the-art results with the performance level of an expert human rater.
Birenbaum A, Greenspan H. Longitudinal Multiple Sclerosis Lesion Segmentation Using Multi-view Convolutional Neural Networks.
In International Workshop on Large-Scale Annotation of Biomedical Data and Expert Label Synthesis 2016 Oct 21 (pp. 58-67).
Springer International Publishing.