Research On Typical Medical Image Segmentation Methods

Medical image segmentation is an important step in medical image processing and analysis,which provides a reliable basis for clinical diagnosis.Brain tissue magnetic resonance image segmentation and tooth panoramic x-ray image segmentation in computer-aided diagnosis technology are widely used medical analysis techniques,which are of great significance for doctors to diagnose diseases,formulate treatment plans,and prepare prognoses.This thesis systematically researches two typical medical image segmentation of brain magnetic resonance images and tooth panoramic x-ray images.The detailed research work is as follows:Firstly,aiming at the problem that the traditional three-dimensional level set methods cannot maintain the temporal consistency of brain tissue segmentation results,this thesis designs a new energy function and proposes a brain tissue segmentation method based on the level set function.A four-dimensional time series image imaging model is established firstly,a variational level set formula with a time consistency constraint on the membership function is introduced,and then the constraint is converted into a vector-valued function.To achieve the purpose of brain tissue segmentation,the membership function is optimized based on the analytically defined mapping.The experimental results show that the method proposed in this thesis has better performance and robustness for MRI segmentation of longitudinal brain tissue,and at the same time effectively improves the temporal consistency of MRI segmentation of longitudinal brain tissue,which has certain practical value.Secondly,in order to solve the problem of fuzzy tooth boundary and tooth root caused by the low contrast and uneven intensity distribution,this thesis proposes a two-stage attention segmentation network on tooth panoramic x-ray images based on deep learning techniques.Firstly,this thesis adopts an attention model which is embedded with global and local attention modules to roughly localize the tooth regions in the first stage.Without any interactive operators,the attention model so constructed can automatically aggregate pixel-wise contextual information and identify coarse tooth boundaries.To better obtain final boundary information,this thesis uses a fully convolutional network in the second stage to further segment the real tooth areas from the attention maps obtained from the first stage.The experimental results show that this method can achieve better results in local detail preservation and edge segmentation,and improve the accuracy of tooth panoramic x-ray image segmentation,which is significantly better than the state-of-the-art methods.