Interactive Segmentation Of Medical CT Images Based On Fully Convolutional Neural Network

In the diagnosis of liver diseases,visualization technology and medical image analysis play a dominant role.By segmenting liver CT images,extracting liver tissue and obtaining corresponding characteristic information,doctors can intuitively understand the details of the patient’s liver,which plays a key role in diagnosis and formulation of the next treatment plan.However,due to poor medical image quality,low contrast,large differences between patients,and irregular appearance caused by pathology,it is difficult for fully automatic segmentation methods to obtain sufficiently accurate and robust results.In order to solve the limitations of automatic segmentation methods for medical image segmentation,this paper proposes two interactive segmentation methods based on fully convolutional neural networks: region growing fully convolutional interactive network and ordinary differential equation fully convolutional interactive network.The region growing full convolutional interactive network improves the growth rules of the region growing method.The gradient value of the pixel under different edge detection operators is calculated as the feature of the pixel,and the pixel feature vector is formed.And put forward a neural network to output the similarity of two pixel feature vectors as the basis for judging whether to grow,and get the preliminary liver segmentation results.The preliminary segmentation result is used as the interactive information of the original image,and the gray channel of the original image is connected together as the input data,and the Region Growing-Fully Convolutional Neural Network(Rg-FCNN)is trained to output the final segmentation result.Compared with other interactive segmentation methods,the method in this paper has simple interactive operation.You only need to select one or two seed growth points in the liver area with the mouse to get the segmentation results.The segmentation results of the experimental test set show that the Dice coefficient reaches 96.69%,and the pixel accuracy rate reaches 99.62%.Compared with the U-net network segmentation results,they are improved by 5.62% and1.06% respectively.The neural ordinary differential equation full convolutional interactive network is composed of two full convolutional neural networks,an automatic segmentation network and an interactive segmentation network.The initial liver segmentation result is generated by the automatic segmentation network,and the user checks whether the segmentation result needs to be modified.If necessary,click on the segmentation error area on the basis of this segmentation to generate the foreground and background interactive information,and use it together with the original image as the input of the interactive segmentation network to obtain the corrected result.For the first time,the automatic segmentation network uses the combination of Ordinary Differential Equation(ODE)and a fully convolutional neural network to segment the liver.Through experimental comparison,choose the position of ODE module to join the U-net network and determine the error tolerance of the module.Adding the ODE module can effectively avoid the problem of network degradation in the deep network.It can also reduce the error tolerance of the ODE module,similar to increasing the depth of the model,so that a deep network can be built even with limited computing resources.model.The experimental test set segmentation results show that through limited interaction,the final segmentation result Dice coefficient reaches 97.11%,and the pixel accuracy rate reaches 99.77%.Compared with the U-net network segmentation result,it is improved by 5.93% and 1.05% respectively.