Multi-scale 3D convolutional neural networks for lesion segmentation in brain MRI
Konstantinos Kamnitsas, L Chen, Christian Ledig, Daniel Rueckert, Ben Glocker · Spiral (Imperial College London) · 2015
We present our 11-layers deep, double-pathway, 3D Convolutional Neural Network, developed for the segmentation of brain lesions.The developed system segments pathology voxel-wise after processing a corresponding multi-modal 3D patch at multiple scales.We demonstrate that it is possible to train such a deep and wide 3D CNN on a small dataset of 28 cases.Our network yields promising results on the task of segmenting ischemic stroke lesions, accomplishing a mean Dice of 64% (66% after postprocessing) on the ISLES 2015 training dataset, ranking among the top entries.Regardless its size, our network is capable of processing a 3D brain volume in 3 minutes, making it applicable to the automated analysis of larger study cohorts.