Uncertainty-Aware Multi-Dimensional Mutual Learning for Brain and Brain Tumor Segmentation

A novel LVPA-UNet network for target volume automatic delineation: An MRI case study of nasopharyngeal carcinoma

“…However, this method might lean towards learning incorrect volumetric structural information due to the resolution difference between the x, y, and z axes in MRIs (image anisotropy), overlooking important details within slices and causing distortion in segmentation results [ 25 , 26 ]. The third approach utilizes hybrid 2D-3D networks, combining the capabilities of 2D networks to focus on fine details within slices with the advantages of 3D networks in understanding the overall spatial structures [ [26] , [27] , [28] ], effectively solving the issue of spatial feature loss encountered with the use of only the 2D networks, as well as the issue of the 3D networks in dealing with the anisotropy of the images [ 29 ].…”