Research And Implementation Of Improved Stacked Cascade Network In Brain Tumor Segmentation

It is well known that deep learning methods have been widely used in the field of medical image processing.Among them,the full convolution network represented by Unet which won the 2015 ISBI Cell Tracking Challenge has been widely used in the field of medical image segmentation.But it does not perform well in dealing with complex medical images(Such as brain MRI).In order to achieve better segmentation performance by adopting the Unet,many researchers have paid more attention on stacking the Unet.However,the stacking process lead to a large increase in the amount of parameters.This is not a good choice when considering the tradeoff between precision and efficiency.Another problem is that as the depth of the network increases,the excessive loss of information is also a tricky problem.To solve these problems,in this thesis,we try to improve the network structure of Unet to make it more suitable for brain tumor segmentation.Firstly,we built a basic block SRNet that is more suitable for stacking,the basic block SRNet is an improvement on the original Unet.It performs only one convolution operation at each level in the encoding or decoding process,and the operation of copy and crop are preserved between encoding and decoding.The main advantage of the SRNet is that the amount of parameters is reduced by 4/5 by comparing with the original Unet.Second,we propose to establish a series of bridge connections between stacked cascaded networks.We try to establish a series of "bridges" inside the stacked cascaded network to make explicit and full use of all available information during the network downsampling They can not only enhance the ability of the network to retain information,but also provide rich information to the subsequent network during forward training,and reduce the risk of gradient disappearance to the greatest extent possible during gradient backhaul.More specifically,some bridge connections will be adopted before the pooling operation in each layer during the downsampling process.It means that each layer in one basic block has a bridge connection with the same feature size from the previous basic block before pooling.At last,we tried to improve the segmentation performance by increasing the network depth or width.We built a network SMCSRNet based on SRNet vertical stacking and bridge connections,and a network SESRNet based on SRNet horizontal stacking.It is worth noting that compared with Stacked RUnet,the performance of our proposed framework SMCSRNet on brain tumor segmentation tasks has been improved,and its network training time costs much less.When further comparing with other state-of-the-art segmentation networks,it can be found that our network test takes less time to segment brain tumors and the performance is as good as the most popular network.Overall,by evaluating the proposed framework on the BRAT2015,it can be proven that the proposed segmentation network has the ability to accurately extract the brain tumor boundary so as to obtain higher recognition quality with high efficiency.