Research On Hippocampal Segmentatio Based On Deep Learning

Now,the world is gradually aging,according to surveys,China’s current Alzheimer’s disease patients about 10 million.For the past 20 years,the mortality rate of Alzheimer’s has skyrocketed to 55% and has become one of the top ten global causes of death,thus denying elderly people a secure and healthy life.The hippocampus is a vital component of the human brain,its placement having an effect on long-term memory,and the amount of atrophy is closely linked to the development of a variety of brain illnesses,such as Alzheimer’s.To accurately diagnose and prevent Alzheimer’s,it is essential to examine the morphological and structural alterations of the hippocampus.However,the boundary contour of the hippocampus is relatively complex,and the difference between the light and dark of each sub-region is not obvious.The challenge of precisely separating the hippocampus from the brain MRI is a major obstacle in the present medical field.Currently,radiology experts are doing most of the laborious segmentation and measurement manually.The rapid advancement of computer science and technology has necessitated the utilization of considerable professional human resources;thus,a plethora of algorithms based on deep learning have been suggested in recent times.Targeting the difficulties of segmentation in medical operations,this was then applied.(1)This paper combines channel attention mechanism with U-Net model.To avert data loss between layers in the U-Net model,and to concentrate more on the goal region,this is the basis of the U-Net model.U-Net has been augmented with a channel attention mechanism and residual module to enhance the algorithm’s precision and diminish the effect of intricate background on the outcomes.A comparison of the SegNet model’s proposed technique and its performance to that of the U-Net model is made.Attention U-Net’s performance surpasses that of both Seg-Net and U-Net models,as evidenced by the results.(2)CE-Net model is incorporated into the encoder-decoder and Inception-Res Net structures,utilizing Attention-Unet to enhance the precision of hippocampus segmentation,thus completing the task.After the encoder,the CE-Net model incorporates the dense hole convolution module(DAC),with its four cascaded branches made up of multi-scale hole convolution,enabling it to capture more extensive and profound semantic features.CE-Net then incorporates residual multicore pooling.Verification of context structure was accomplished through the ablation experiments of DAC and RMP modules.The model’s precision can be greatly enhanced through the combination of DAC and RMP modules with an encoder-decoder network structure.(3)Attention-Unet and CE-Net models have poor segmentation effect on the hippocampus,which has more complicated edge information,a novel hippocampal segmentation algorithm based on UNet + + model and Conv Next backbone network is proposed.This paper employs Conv Next as the encoder to capture the high-level semantic information,which the traditional network model is unable to effectively do.U-Net++’s up-sampling is employed as a decoder module to join together the multi-level semantic information obtained from the encoder.The network depth is shallower than the original U-Net++ model.This design ensures the accuracy of the model and reduces the operation cost.Compared with other models,results demonstrate that Conv Next’s model can augment segmentation precision and enhance the practicality of medical segmentation.The performance of the algorithm,evaluated by the similarity coefficient,precision rate and recall rate of Dice,is demonstrated in this paper through the training of the proposed network on an ADNI data set.