Research On Target Segmentation Of Radiotherapy For ESCC And Its Standard Service Interface

Esophageal Squamous Cell Carcinoma(ESCC)is a high incidence rate of cancer,and China is also a country with multiple esophageal squamous cell carcinoma.At present,esophageal squamous cell carcinoma has become one of the main causes of death in China.With the development of medical technology,radiotherapy has become one of the main treatment methods of esophageal squamous cell carcinoma,and radiotherapy needs to use the images obtained by computed tomography(CT)technology to make treatment plans.Therefore,it is the main basis for doctors to make radiotherapy plan to delineate radiotherapy target on CT images.It takes a lot of time to segment tumor manually,and the segmentation effect is affected by doctors' experience,so it is necessary to achieve automatic target delineation to avoid these problems.However,it is very difficult for the computer to automatically delineate the target area of radiotherapy because of the variety of shape,location,structure,surrounding environment and different cases of esophageal tumor.In view of the above problems and background,this thesis,with the support of Shandong provincial key research and development project "key technology of individualized radical synchronous radiotherapy and chemotherapy based on multi center clinical research queue",studies a multi-depth neural network fusion method for esophageal squamous cell carcinoma target area delineation with the image data of 50 cases of radiotherapy from Shandong cancer prevention and Treatment Research Institute as the research object.On the basis of this,a computing service platform is constructed,which takes the automatic target delineation of esophageal squamous cell carcinoma radiotherapy as an example.The main work is as follows:(1)This thesis proposes a new CT image preprocessing method for esophageal squamous cell carcinoma tumor target delineation.In order to solve the problem of incomplete information and uneven gray scale in linear conversion in CT window reading method,a new CT image reading algorithm based on pixel density is proposed.The algorithm uses the pixel density of the original data to map to the adaptive length interval.In theory,the algorithm ensures that the gray level of the processed image is uniform and the information is saved completely.In view of the problem that the input image is too large to affect the training efficiency and accuracy of the model,this thesis proposes a method to locate the esophagus and cut out the esophageal region with the help of spine and double lungs.In this method,the characteristic spine and lung in CT images are used to locate the esophageal region effectively,limit the input size of the model,and improve the training efficiency and accuracy of the model.(2)A fusion method of multi model target area is proposed.In this thesis,two kinds of deep learning models FCN and u-net are used to segment esophageal tumor independently,and then a linear fusion algorithm is proposed to fuse the sketch results of different models.Through the quantitative analysis of the two segmentation results,the algorithm calculates the consistency between each segmentation and selects the fusion mode.For the data with poor consistency,this thesis proposes a linear fusion algorithm,which uses pixel interpolation to find the corresponding relationship of each contour and get the fused contour.In the first mock exam,the method can effectively solve the over segmentation and under segmentation problem of single model.The experimental results show that the fusion result is 1.5% higher than FCN and 3.4% higher than u-net in Dice coefficient.(3)To design and realize the calculation service of esophageal squamous cell car cinoma radiotherapy target delineation.In this thesis,websocket technology,combined with the previous research results,based on Python design and implementation of a simple and stable architecture,good transplantation of esophageal cancer segmentation computing services.The service has the ability of real-time information transmission and parallel processing.(4)Build user oriented computing service platform.The computing service platform provides researchers with data access interface,file upload and download services and application access services,and provides medical staff with data display,patient course query and other services.The platform has designed a set of criteria.Researchers can make their own research content into services and publish it to the computing platform for other users to use.To sum up,this thesis studies four key issues of medical image preprocessing,medical image segmentation,computing services and platform building,and puts forward corresponding effective methods.These studies have a great role in promoting the realization of automatic segmentation of esophageal squamous cell carcinoma radiotherapy target.