Research On Low Resolution PET Image Segmentation Algorithm And Software Design

As one of the most advanced medical imaging technologies,PET can display the information of human metabolic function at the molecular level without intervention,which provides the necessary basis for doctors’ clinical auxiliary diagnosis and treatment.At present,the image resolution of the PET instrument still used in clinic is low.Due to the high cost of updating hardware equipment,it is a low-cost and more convenient way to improve the image quality through the improvement of software algorithm.As the basis of image recognition and classification,medical image segmentation aims to extract the region of interest in the image.The segmentation results provide data basis for image attenuation correction and improve the quality of reconstructed image.Therefore,accurate extraction of the target region is to guarantee the later medical image analysis.There are some problems in head PET images,such as star-shaped artifacts,edges deficency and low resolution.To solve the above problems,a CV model with oval as the prior shape has been proposed.Firstly,the three-dimensional head fault has been projected in the direction of coronal plane and sagittal plane to obtain the two-dimensional projection map information.Then,the projection image has been re-valued by OTSU algorithm,and the parameters of the oval have been obtained by using the geometric features of the edges.The oval has been used as the prior shape of the segmentation model by fitting the parameters.At the same time,the fitted ellipse has been used as the initial contour to reduce the time for the curve to evolve to the target location.At last,the symbol distance function of the prior shape has been added to the CV model through the shape comparison function to control the curve evolution.The experimental results show that the contour segmentation of head PET image with poor quality is effective.Pointing at the problems in body part PET image,such as uneven distribution of gray scale,artifact and missing edge,an algorithm combining the global information of CV model with the local information of DRLSE model has been proposed.The least square method has been used to fit the ellipse as the shape constraint,and theevolution of the curve has been controlled by the combination of the image information and the external constraints.The experimental results show that the proposed algorithm can segment the contour correctly when dealing with body PET images with uneven distribution of gray scale,artifacts,and missing edges.In order to facilitate the application of PET image segmentation results,a medical image segmentation software has been developed on the MATLAB GUI platform.The software includes window level adjustment function and window width adjustment function,header file information display and other auxiliary functions.Combining with the function of contour segmentation of PET image realized by the algorithm in this paper,a friendly operation interface has been provided for users.