Problem
=========
Brain MRI segmentation using adversarial training approach

Dataset
======
55 T1 weighted brain MR images (35 adults and 20 elders) with respective label maps.


Contributions
==========
1. The authors suggest an adversarial loss in addition to the traditional loss.
2. The authors compare 2 Generator (Segmentor) models - Fully convolutional and dilated networks.

https://i.imgur.com/orhWhoM.png

Dilated network
------------------
Using conv layers, allows for larger receptive field with fewer trainable weights (compared to the FCN option).


However, the authors claim the adversarial loss contributes more when applying the FCN model

This work aims to produce more spatially consistent MRI image when the patient is breathing during MRI acquisition. 

https://i.imgur.com/wWMQa1D.png

This work improves the performance of a segmentation network by utilizing unlabelled data. They use a discriminator (they call EN) to distinguish between annotated and unannotated examples. They then train the segmentation generator (they call SN) based on what will fool the discriminator. 

https://i.imgur.com/7CfKnh5.png

Three training phases are shown above

This work is really great. They are using the segmentation to condition the discriminator which will learn to point out flaws when applying the segmentation to the unlabelled examples. Then these flaws in the segmentation are corrected by using the gradients from the discriminator to adjust the segmentation.

In contrast with other semi-supervised approaches which learn a latent space for all samples, labelled and unlabelled, and then uses this space to learn a classifier or segmentation; this approach looks for the boundaries of the space only. The unlabelled examples are used to bias the representation learned by the segmentation network to conform to the distribution represented by all observed examples.

Read this paper for more: https://arxiv.org/abs/1611.08408

Poster:
https://i.imgur.com/eR5jgwn.png

The goal of this work is to classify histopathology images into benign and malignant.  They use the BreaKHis and IICBU 2008 lymphoma datasets.

They use a VGG network for feature extraction from each image. Then on these VGG feature vectors they learn [Fisher Vectors ](https://prateekvjoshi.com/2014/08/23/image-classification-using-fisher-vectors/) which they use to make a prediction.

It is unclear why Fisher Vectors are more useful than the fully connected layers of the VGG net that they replace. It is not clear how much analysis was performed for the VGG baseline. Also, as a baseline a VGG network should have been trained from scratch to extract domain specific features. 

Poster:
https://i.imgur.com/fgzmeYv.png

The authors state that the usual approach to cope with few training samples is data augmentation. They extend a method of modelling the data from \cite{10.1016/j.media.2017.02.003} and use it to train a neural network. The figure below shows the overview:

https://i.imgur.com/joLNyfc.png

At the core of deformation model they determine a set of $m$ landmarks $s_i$ which they will deform and then perform an affine transformation to warp the image to align to these points. The points are moved in a constrained way. They state the constraint is a "multi-level B-spline scattered data approximation".

Here is the poster: https://i.imgur.com/enQQqxC.png