Research And Realization Of Segmentation Algorithm Of Brain MR Images Based On Deep Learning

Many diseases of the brain to brain gray and white matter off the relationship,such as leukoaraiosis,cerebral gray matter heterotopia,stroke,intracranial tumor,cerebral hemorrhage and other diseases.Magnetic resonance imaging of the brain has the advantages of convenient and non-invasive advantages for diagnosis with higher in intracranial lesions,but each sequence of brain magnetic images need several physicians view,which increases the workload of doctors,it is likely to affect the best treatment period patients.Therefore,it is necessary for medical workers to study the computer aided technology of brain magnetic resonance image segmentation of gray matter,and it is of great significance to achieve fast and accurate segmentation of gray matter region.The segmentation of brain magnetic resonance images based on deep learning is divided into two steps,modeling and segmentation.First,based on the depth of learning algorithm to model.Which mainly includes the pretreatment of nuclear magnetic resonance image and the feature modeling of deep learning.The pretreatment stage,after the brain MRI image denoising and image enhancement,because the skull belonged to highlight regions,respectively from the brain MRI images on both sides of the skull to remove the vertical middle traversal,the pre segmentation regions,taking into account the deep learning of the advantages of big data processing,the study of mapping individual pixels to 16 pixels around stage;feature modeling based on deep learning,with deep learning deep strong learning ability,feature extraction of pre segmentation regions,respectively using SAE and CNN deep learning two learning methods.Secondly,the segmentation stage of the brain gray matter,the first FCM made a rough segmentation,clustering results come to gray and white matter of the two categories.Because the FCM clustering result is not accurate,remove some points in the vicinity of two kinds of cluster centers,SVM trained classification parameters as classifier,feature extraction of the original deep learning to do a adaptive classification,obtained the ideal gray segmentation results.In the evaluation of the experimental results,the feature vectors of the image are extracted and the results are evaluated by the gray level co-occurrence matrix.After the experiment,two kinds of algorithms have achieved the ideal segmentation results,but the CNN of the indicators are slightly better than SAE.Among the parameter setting and the algorithm efficiency,CNN is slightly better than SAE.Therefore,it is believed that the CNN algorithm is better than the SAE algorithm in the segmentation of brain gray matter.In summary,the brain MRI image deep learning SAE and CNN two algorithm of segmentation based on the deep learning feature modeling of region of interest in brain MRI image,feature classification based on FCM and SVM,to improve the classification performance and efficiency.