Xiao Liu and Spyridon Thermos
Thu 02 Dec 2021, 13:00 - 14:00
Online Teams

If you have a question about this talk, please contact: Mehrdad Yaghoobi Vaighan (myvaigha)

Talk1: Despite recent progress in medical image analysis with deep learning and machine learning, inference performance on new datasets, acquired from distinct scanners or clinical centres, is known to significantly decrease. In this presentation, I will present how we decompose the data into disentangled representations with deep convolutional neural networks and how to efficiently use or learn the representations for robust medical image analysis. In particular, we use meta-learning to learn generalised disentangled representations, and disentanglement allows us to train the model with unlabeled data. The paper is available at https://link.springer.com/chapter/10.1007/978-3-030-87196-3_29.

Talk 2:This work proposes a data augmentation model to synthesize cardiac images, mixing and matching anatomy factors of variation that are already captured in existing datasets. The work was inspired by a process called vector arithmetic, which is well known in the vision domain. It differs from a traditional synthesis model in that it does not sample from a noise distribution, instead of extracted or encoded factors of variation of an image are combined to generate an intermediate representation that preserves the fidelity of the encoded factors. The paper is available at https://link.springer.com/chapter/10.1007/978-3-030-87199-4_15.

This talk is included in these lists:

Note that ex-directory lists are not shown.