Deep Learning-Based Automated Segmentation of Brain and Vertebral Substructures for Radiotherapy in Pediatric Medulloblastoma

To evaluate nnU-Net and FuseNet for automatic segmentation of pediatric medulloblastoma substructures, 60 radiotherapy patients (≤5 and >5 years) were analyzed. Brain substructures were delineated on CT-MRI fusion images, vertebral ones on CT. U-Net, nnU-Net and FuseNet were trained (24 cases/group) and tested/validated (6 cases/group), with 20 external cases verifying generalization. DSC was compared with Atlas; nnU-Net/FuseNet’s HD95, RAVD and manual correction time were evaluated. FuseNet performed best in brain segmentation, outperforming Atlas and U-Net in vertebral substructures for both age groups (P=0.028/0.005 and 0.005/0.005) but not differing from nnU-Net (P=0.107/0.236). Its DSC exceeded 0.8 for most substructures (except cerebellar anterior lobe/hippocampus in ≤5 years; hippocampus in >5 years), with shortest correction time. nnU-Net achieved good segmentation; FuseNet improved brain segmentation accuracy via dynamic multimodal feature fusion, with highest correction efficiency.