Research On Segmentation And 3D Reconstruction Of Pituitary Tumor Based On MRI Images

In recent years,social and economic development has brought great improvements to people’s standard of living in all aspects,but it has also disrupted people’s former work and life routine.Brain tumours,a disease of the brain with a high mortality rate,are gradually moving to the younger generation.Pituitary tumours,a common type of brain tumour in the intracranial saddle area.Magnetic resonance imaging（MRI）technology is widely used for clinical medical imaging of the brain due to its high quality,high resolution and multimodality.However,the question of how to segment the pituitary tumour portion of the obtained MRI images of pituitary tumours has always been a difficult issue in clinical diagnosis and treatment.The common method today is to manually label the pituitary tumour by the appropriate specialist,which is time consuming and may be biased,so it is a challenge to automatically and accurately segment the pituitary tumour from the MRI image using computer technology.On the other hand,a single segmented image of a pituitary tumour contains only two-dimensional information about the lesion,the three-dimensional structure of the pituitary tumour still has to be imagined by the physician.Therefore,realistic research on the segmentation and 3D reconstruction of pituitary tumours based on MRI images still holds great promise for clinical paramedical applications.This thesis proposes a solution that can be applied to the segmentation and 3D reconstruction of pituitary tumour MRI images with high segmentation and reconstruction results.By taking into account the local information of the LBF model,the energy function of the threshold level set is improved,and the segmentation results of the 3D confidence connection or threshold level set methods are morphologically processed as the initial contour Mask of the improved Level Set model,which solves the problem that the LBF model is sensitive to the selection of the initial contour and the problem that the 3D confidence connection and threshold level set methods are difficult to segment pituitary tumours with weak The problem that the LBF model is sensitive to the initial contour selection and the 3D confidence linkage and threshold level set methods are difficult to segment the weak edges and internal inhomogeneities of pituitary tumours.Experiments showed that the improved Level Set model improved the segmentation accuracy of brain microadenoma by nearly 10 percentage points.On the other hand,by subjecting the segmented pituitary tumour MR slice series in this thesis to face mapping based on the MC method and body mapping based on the RC method,the reconstructed model was also able to achieve a volume overlap of more than 90% with the model constructed from the gold standard slices basically.（1）The original MRI brain pituitary tumour series were first pre-processed,including linear unification of the images,grey scale transformation,and image edge preservation filtering,to provide the basis for the subsequent coarse segmentation.（2）Coarse segmentation is performed by segmenting the results obtained from pre-processing based on 3D confidence linkage and threshold-based level set algorithm to obtain the approximate shape and contour of the pituitary tumour,and optimizing the segmentation results by performing mathematical morphological processing,etc.on the coarse segmentation results,and using them as the initial Mask for fine segmentation.（3）Fine segmentation is performed by using the optimized result image as its initial Mask and performing a fine segmentation based on the improved Level Set model to obtain a series of images of the pituitary tumour for precise segmentation.（4）Finally,the segmented pituitary tumour series images are reconstructed and visualised in 3D using MC-based face painting and RC-based body painting methods,and the final 3D pituitary tumour model in STL format is exported for clinical application.