Representation Learning for Non-Melanoma Skin Cancer using a Latent Autoencoder

TL;DR

This paper explores combining autoencoders and adversarial autoencoders to improve image reconstruction and representation learning for non-melanoma skin cancer histological images, achieving better FID scores and realistic image interpolations.

Contribution

It introduces a novel combination of autoencoders and adversarial training for histological image representation learning, enhancing reconstruction quality and interpretability.

Findings

FID scores improved from 76 to 50 with adversarial training

Representation benchmarks increased by up to 3%

First demonstration of smooth, realistic image interpolations in this context

Abstract

Generative learning is a powerful tool for representation learning, and shows particular promise for problems in biomedical imaging. However, in this context, sampling from the distribution is secondary to finding representations of real images, which often come with labels and explicitly represent the content and quality of the target distribution. It remains difficult to faithfully reconstruct images from generative models, particularly those as complex as histological images. In this work, two existing methods (autoencoders and adversarial latent autoencoders) are combined in attempt to improve our ability to encode and decode real images of non-melanoma skin cancer, specifically intra-epidermal carcinoma (IEC). Utilising a dataset of high-quality images of IEC (256 x 256), this work assesses the result of both image reconstruction quality and representation learning. It is shown that adversarial training can improve baseline FID scores from 76 to 50, and that benchmarks on representation learning can be improved by up to 3%. Smooth and realistic interpolations of the variation in the morphological structure are also presented for the first time, positioning representation learning as a promising direction in the context of computational pathology.