Computer Vision News - February 2017

DeepMedic , a Deep Learning method for tumors segmentation for 3D medical images, provides software for brain lesion segmentation based on a multi-scale 3D Deep Convolutional Neural Network coupled with a 3D fully connected Conditional Random Field . DeepMedic is an award-winning system, winner of the ISLES 2015 competition on traumatic brain injury (TBI) segmentation tasks. DeepMedic software comes with the specific network configuration that won the challenge, with the addition of being fully configurable, so that you can set up your own networks to your own dataset and needs. The software is available on Github . Just to give you a taste of the difficulty of the tasks, you can observe below how different in size and distribution TBIs can be (Fig. 1), and the significant overlap in intensity between the distributions of healthy and non-healthy brain tissue (Fig. 2). Fig. 1. Wide spatial distribution of lesions Fig. 2. Healthy (green) and injured (red) tissue. DeepMedic comes with a proposed architecture of a three-dimensional CNN with dual pathway and 11-layers. 8 Computer Vision News We Tried for You DeepMedic - Deep Learning method for tumors segmentation for 3D medical images Tried for You