Robust Classification of Histology Images Exploiting Adversarial Auto Encoders

Deep learning (DL) thrives on the availability of a large number of high quality images with reliable labels. Due to the large size of whole slide images in digital pathology, patches of manageable size are often mined for use in DL models. These patches are variable in quality, weakly supervised, individually less informative, and noisily labelled. To improve classification accuracy even with these noisy inputs and labels in histopathology, we propose a novel method for robust feature generation using an adversarial autoencoder (AAE). We utilize the likelihood of the features in the latent space of AAE as a criterion to weigh the training samples. We propose different weighting schemes for our framework and evaluate the effectiveness of our methods on the publically available BreakHis and BACH histopathology datasets. We observe consistent improvement in AUC scores using our methods, and conclude that robust supervision strategies should be further explored for computational pathology.

I. INTRODUCTION

Supervised deep learning (DL) models have consistently shown promising results for automated image analysis in medical image analysis over the last eight years [1], [2]. However, the success of DL models depends on the availability of large datasets of high quality and correctly labelled images for training. When the quality of the training images or the accuracy of their labels degrade, the accuracy of the DL models trained using them reduces drastically [3]. Consequently, for automated medical image analysis in general, and computational pathology in particular, medical experts on a research team need to carefully label, annotate, and curate whole slide images (WSIs) to prepare training and testing datasets. This process often involves precise annotations of regions of interest (ROIs) so that high quality and homogeneous patches (sub-images) of anatomical structures can be mined. These patches then inherit the same label for as the ROI from which these are mined. This data preparation process is time-consuming and expensive.