CLASS-M: Adaptive stain separation-based contrastive learning with pseudo-labeling for histopathological image classification

TL;DR

A semi-supervised patch-level histopathological image classification model, named CLASS-M, that does not require extensively labeled datasets and is compared with other state-of-the-art models on two clear cell renal cell carcinoma datasets.

Abstract

Histopathological image classification is an important task in medical image analysis. Recent approaches generally rely on weakly supervised learning due to the ease of acquiring case-level labels from pathology reports. However, patch-level classification is preferable in applications where only a limited number of cases are available or when local prediction accuracy is critical. On the other hand, acquiring extensive datasets with localized labels for training is not feasible. In this paper, we propose a semi-supervised patch-level histopathological image classification model, named CLASS-M, that does not require extensively labeled datasets. CLASS-M is formed by two main parts: a contrastive learning module that uses separated Hematoxylin and Eosin images generated through an adaptive stain separation process, and a module with pseudo-labels using MixUp. We compare our model with other state-of-the-art models on two clear cell renal cell carcinoma datasets. We demonstrate that our CLASS-M model has the best performance on both datasets. Our code is available at github.com/BzhangURU/Paper_CLASS-M/tree/main

Figures (6)

• Figure 1: The block diagram of our Contrastive Learning with Adaptive Stain Separation and MixUp (CLASS-M) for semi-supervised histopathological image classification. Orange part shows adaptive stain separation, where OD means Optical Density space. Green part shows mixup on both labeled and unlabeled samples.
• Figure 2: Contrastive learning on Hematoxyling images and Eosin images
• Figure 3: Examples of tiles with the size of 400$\times$400 from (a) Utah ccRCC dataset (in 10X) and (b) TCGA ccRCC dataset (in 20X).
• Figure 4: Mapping pixels onto the 2D OD space
• Figure 5: Visualization of our CLASS-M model's predictions on Utah ccRCC WSIs from test set. The polygons show the annotations on WSIs. (Green: Normal/Benign, Yellow: Low risk cancer, Red: High risk cancer, Dark Grey: Necrosis) Each polygon labeled as low/high risk cancer marks the region of a certain cancer growth pattern. The heatmaps show the model's predictions on all foreground tiles when the predictions are not Normal/Benign. The strength of color shows the confidence of prediction.