Computer Aided Diagnosis And Treatment System For Nasopharyngeal Carcinoma Based On Multi-Modality Medical Image

As a category of common malignant tumors in clinic,head-and-neck cancers which lead to high mortality are common malignant tumors.Nasopharyngeal carcinoma(NPC)which is one of the most common headand-neck cancers poses a serious threat to human life.Clinical studies have shown that early detection,accurate diagnosis and treatment of NPC can significantly improve the cure rate and prolong the survival period of patients.To gain useful information for proper diagnosis and treatment,in general,radiologists or other specialists have to read a large number of medical images and manually identify tumor regions slice by slice.However,it is an extremely tedious and time-consuming procedure highly depending on expertise and experience of radiologists.Therefore,on the basis of in-depth analysis of the process of diagnosis and treatment of NPC,a computer aided diagnosis and treatment system for NPC based on multimodality medical image is proposed.The system mainly focuses on assistant diagnosis of NPC,automatic segmentation of tumor area,automatic segmentation of tumor sub-regions and 3D visualization of NPC.In the early screening stage of NPC,CT scanning is normally employed in NPC diagnosis.Therefore,this paper proposes a diagnostic method of NPC based on CT image.Firstly,the segmentation of nasopharyngeal region is realized based on adaptive template method,and the redundant information of non-nasopharyngeal region image is removed.Then,a deep convolution neural network is proposed to realize the classification of normal and abnormal two-dimensional CT slice images.So as to judge whether an individual has NPC or not and determine the possible slice location of tumors.After the diagnosis of NPC,an experienced radiologist usually delineates the tumor contour in MR images which is a crucial issue in radiotherapy planning.Therefore,a NPC segmentation algorithm based on multi-modality MRI fusion network is proposed.On the basis of improving the existing network structure,the idea of multi-modality fusion and a training strategy named self-transfer are proposed in this network.In this paper,we compare the segmentation results based on different methods.Experiments show the effectiveness of the proposed segmentation algorithm for NPC.In order to formulate a reasonable radiotherapy plan and improve the local control rate of NPC,this paper proposes a sub-region segmentation algorithm for NPC based on CT and MR image registration.Firstly,a linear registration method based on mutual information is used to realize the spatial registration of CT and MR images of the same patient.And then the segmentation results of tumor regions are mapped to CT images based on the same transformation matrix.Additionally,an adaptive threshold based segmentation algorithm is proposed to further divide the tumor regions into three sub-target regions with different malignant degrees,which is especially relevant in radiotherapy planning.In clinical diagnosis,CT and MR images can provide different information for radiologists.Therefore,this paper realizes the fusion of CT and MR images based on linear weighted fusion algorithm which can display the information of different modalities in the same image.Then this paper realizes three-dimensional visualization of head structure and tumor area based on surface rendering technology,which is convenient for radiologists to diagnose the disease.This system integrates assistant diagnosis,treatment and 3D visualization of NPC.On the one hand,the automatic diagnosis of NPC and the automatic segmentation of tumor area can alleviate radiologists' workload.On the other hand,the automatic segmentation of NPC subregions and 3D visualization can provide a reliable basis for radiologists to analysis patients' condition and formulate treatment plans.Therefore,the system implemented in this paper has certain application value in clinic.