Prostate Cancer Segmentation using Manifold Mixup U-Net
- May. 2019
- by Wonmo Jung et. al.
The scarcity of labeled data is a challenging problem in medical segmentation. Here, we suggest to apply manifold mixup, a recently proposed simple regularizer that utilizes linear combinations of hidden representations of training examples, on prostate cancer segmentation using MR image. Manifold mixup applied to either the encoder or decoder outperformed training without mixup and mixup applied on the input space.
Wonmo Jung, Sejin Park, Kyu-Hwan Jung and Sung Il Hwang