Combining Deep Neural Network For The Algorithm Study Of Hippocampus Segmentation

The state of the brain’s hippocampus plays a key role in the diagnosis of neurodegenerative diseases such as Alzheimer’s disease.Changes in the size and shape of the brain’s hippocampus are early indicators of Alzheimer’s disease,and MRI is the main way to diagnose brain lesions.Accurate and efficient segmentation of the brain hippocampus from MR images of human brain can improve the accuracy and speed of doctors’diagnosis.Although there have been many studies focus on the automatic segmentation of the hippocampus,but as a result of MR imaging in low contrast between the hippocampus and other organizations,and MR imaging of the brain itself there is a lot of noise,so most of the result of the hippocampus segmentation algorithm accuracy is low,can’t meet the requirements of clinical performance of the computer aided diagnosis.In order to obtain the segmentation results with clinical application value,this paper focuses on the research of accurate and efficient automatic segmentation algorithm of brain hippocampus tissue.In this paper,we propose an algorithm based on deep neural network,which includes two stages.Firstly,the brain hippocampus was initially segmented based on the deep neural network.After the clipping and preprocessing of the sample data was completed,the brain hippocampus region was expanded by means of geometric transformation.Then,the improved U-NET deep neural network model was utilized to achieve the initial segmentation of the brain hippocampus;On this basis,the initial segmentation results obtained from the deep neural network model were combined with the original BRAIN MR images to obtain the initial contour of the evolution of the level set curve.The local binary fitting model based on the level set calculation method was used to obtain the precise segmentation results of the brain hippocampus through the evolution of the level set curve.In this paper,the depth neural network is combined with the level set model,and the initial segmentation results are taken as the initial contour of the level set curve,which effectively overcomes the problems of the traditional level set model,such as sensitivity to the initial contour and high computational cost,etc.,and the segmentation results with higher precision can be obtained.On the basis of the above theoretical research results,a brain horse-body segmentation prototype system was developed and realized,which can be used for clinical auxiliary diagnosis and further algorithm research.In this paper,through comparative experiments,the proposed algorithm is compared with other classical level set models and other algorithms.The experimental results show that the proposed algorithm can achieve rapid and accurate segmentation of the brain’s hippocampus.