SepVAE: a contrastive VAE to separate pathological patterns from healthy ones

TL;DR

SepVAE introduces a novel contrastive variational auto-encoder that effectively separates pathological patterns from healthy ones by disentangling salient and common features, improving interpretability and performance in medical imaging.

Contribution

The paper proposes two new regularization losses for CA-VAEs, enhancing the separation of salient features and preventing information sharing between latent spaces.

Findings

Outperforms previous CA-VAEs on medical datasets

Achieves better disentanglement of pathological and healthy features

Demonstrates effectiveness on natural images like CelebA

Abstract

Contrastive Analysis VAE (CA-VAEs) is a family of Variational auto-encoders (VAEs) that aims at separating the common factors of variation between a background dataset (BG) (i.e., healthy subjects) and a target dataset (TG) (i.e., patients) from the ones that only exist in the target dataset. To do so, these methods separate the latent space into a set of salient features (i.e., proper to the target dataset) and a set of common features (i.e., exist in both datasets). Currently, all models fail to prevent the sharing of information between latent spaces effectively and to capture all salient factors of variation. To this end, we introduce two crucial regularization losses: a disentangling term between common and salient representations and a classification term between background and target samples in the salient space. We show a better performance than previous CA-VAEs methods on three medical applications and a natural images dataset (CelebA). Code and datasets are available on GitHub https://github.com/neurospin-projects/2023_rlouiset_sepvae.