Knowledge distillation on individual vertebrae segmentation exploiting 3D U-Net

Luís Serrador, Francesca Pia Villani, Sara Moccia, Cristina P. Santos

Abstract: Highlights•Knowledge distillation is applied for individual vertebrae segmentation on CT scans.•Resampling and memory cleaning techniques improve performance of neural networks.•Knowledge distillation helps balance performance and network size•Number of parameters and inference time reduced without compromising performance.•High accuracy with low computational resources, allowing deployment on medical devices.