Brain Tumor Slice-level Image Segmentation And 3D Reconstruction Algorithm Research

With the gradual increasing incidence of glioma,medical image processing as a key auxiliary diagnosis and treatment technique has received more and more attention.Thereinto,the organic combination of brain magnetic resonance image(MRI)segmentation and reconstruction algorithms greatly facilitates growing medical tasks,e.g.,diagnosis and treatment decision-making,simulated surgery,postoperative planning,and medical education.Therefore,the design of segmentation and reconstruction algorithms for MRI has always been an active research hotspot in medical image processing and computer-aided diagnosis.However,Gaussian noise and Rayleigh noise are often superimposed during the imaging process,causing the degradation of image quality,further resulting in limited segmentation accuracy.Meanwhile,the existing automatic or semi-automatic segmentation methods are generally incapable of accurately predicting the boundary of the region of interest,which makes it difficult to present the complex nonlinear surfaces of the intracranial tissue during reconstruction.To address the above challenges,this thesis analyzed the self-portrait of the MRI and studied the existing segmentation and reconstruction methods in detail.Novel brain magnetic resonance image segmentation and reconstruction algorithms are proposed to realize 2D slice-level segmentation and 3D visualization reconstruction of brain lesion areas.The main contributions of this thesis are summarized as follows:(1)To address the limited accuracy of the existing segmentation methods for MRI,this thesis proposed a novel neural network-based segmentation model,marked Active Contour Unet(ACU-net).Technically,the proposed model firstly makes full utilization of the imaging characteristics of T1-weighted images,T2-weighted images,contrast-enhanced T1 sequences,and water-suppressed imaging technology.The spatial and appearance correlations of the convolutional layer are mapped through depthwise separable convolutions.Then,the dense residual module is introduced to realize more comprehensive captures of abnormal regions in brain tumor images.In addition,the active contour constraints are employed to solve the puzzle of blurred peritumoral edema areas and achieve the efficient segmentation of the enhanced necrotic tumor and the core tumor regions in the lesion areas.Finally,extensive ablations and comparisons verified the effectiveness and superiority of the proposed method.(2)To address the problem that the existing 3D reconstruction algorithms of MRI are hard to deal with the change of slice curvature and the discontinuity of the surface,this thesis proposed a stacked reconstruction method after image segmentation.Technically,the proposed method makes full use of the small-area tumor block slice to construct a spatial coordinate system.The bicubic interpolation is employed to augment the pixels located in the horizontal and vertical columns while integrating the stacked sequences to extract isosurface to approximate the surface of the reconstructed volume.Then,the Phong local illumination model is introduced to superimpose different reflected light to render the image smoothness and RGB values for the reconstruction of the panoramic view of small-area tumors.Finally,the effectiveness of the proposed method is verified by competing experiments and ablations.