Brain Tissue Image Segmentation By Deep Learning

With the development and improvement of various medical imaging technologies,the medical imaging,providing the scientific foundation for disease analysis,has been playing an important role in medical diagnoses.Because of efficient information collection,the medical images have characteristics of higher resolution and multi-function.So it is attracted a variety attentions about how to explore valuable data hidden in medical images to assist doctors make diagnoses precisely and efficiently.As a branch of machine learning,deep learning has been fashionable in the past ten years,and been widely applied in artificial intelligence in recent years,for example,it shocked the world that AlphaGo challenged and won Lee Sedol in the March of this year.In the last several years,various excellent network architectures were applied in image classifications and segmentations,and achieved outstanding results,but there is not any single architecture fitting for all kinds of these problems.This paper improved the fully convolutional network,and applied the improved networks to complete the sematic segmentation of brain images collected by MRI and neuron images collected by electron microscope,and obtained excellent results.First,this paper describes the development of MRI and deep learning,and shows the importance of MR brain image segmentation.Furthermore,in order to make the segmentation results more precise,this paper improves and optimizes the existing fully convolutional network's performance using Caffe,the deep learning framework,and also evaluates the segmentation results.Finally,the optimized network is used to segment the neuron images,in order to get the bound of neurons.