Computer Vision News - September 2020

Research 6 One interesting aspect of this method is that even though it’s based on an adversarial approach, it doesn’t rely on generative networks (GANs) but instead it offers explainability with parameters that are relevant toMR image segmentation, including various commonly seen differences in image appearance. An example is shown in Fig. 1. Realistic Adversarial Data Augmentation for MR Image Segmentation Hi again! Thepaper presented thismonth is about amethod of data augmentation for medical image segmentation, which is a great continuation of the previous’ month topic! The first author is Chen Chen from the group of Daniel Rueckert at Imperial College London. The network presented in this work attempts to offer a method that can help deep neural segmentation to be more robust and better generalise. by Ioannis Valasakis (@wizofe) Figure 1. An example of sub-volume sampling of a 3D MRI use for data augmentation. Image courtesy of AISummer.com. Medical diagnostic tasks, such as the prediction and segmentation of tumours, including lung, liver, and breast scans are using often deep learning models for