Applying Disentanglement in the Medical Domain: An Introduction for the MAD Workshop

Abstract

AbstractFor medical applications, trustworthiness, interpretability, and robustness are necessary properties of (deep) neural networks. For generative models, one approach towards this could be analyzing and structuring the latent space representation. In this context, the term disentanglement is often used, but still not uniquely defined. In 2022, we organized the first workshop about Medical Applications with Disentanglements (MAD) at the MICCAI conference in Singapore (https://mad.ikim.nrw/). The workshop had a general call for disentanglement papers in the medical field and the submitted papers are published in a Springer challenge proceedings. The aim of the introduction paper of this proceeding is to present the necessary background information for them. Thus, we give a short overview of this field and how challenges for deep learning in healthcare could be addressed with the help of disentanglement.KeywordsDisentanglementGenerative modelsDeep learningMAD WorkshopMICCAI