Improved Medical Image Segmentation Method and Three-Dimensional Reconstruction Based on 3D-Unet

In the realm of medicine, the incorporation of deep learning methods for visualizing lesion regions in medical images holds the potential to alleviate the diagnostic burden on clinical practitioners. Within this context, medical image segmentation, as a pivotal component of three-dimensional (3D) reconstruction, assumes pronounced significance due to its capacity for accurately delineating lesion areas. This paper introduces an enhanced medical image segmentation approach based on the 3D-Unet network architecture, with the aim of achieving high-quality three-dimensional reconstruction. This method introduces a Convolutional Block Attention Module (CBAM) attention module into the conventional 3D-Unet structure to enhance feature representation, while also integrating Pixel Attention and residual connections to further optimize model performance. Through refined model training experiments, notable improvements in image segmentation accuracy are achieved. Additionally, successful 3D reconstruction of brain tumor segmentation images and cerebral images using the Monte Carlo (MC) Algorithm and light projection algorithm is achieved, which provides richer information for medical image visualisation research.