H&E Multi-Laboratory Staining Variance Exploration with Machine Learning

Improved accuracy in colorectal cancer tissue decomposition through refinement of established deep learning solutions

Hematoxylin and eosin-stained biopsy slides are regularly available for colorectal cancer patients. These slides are often not used to define objective biomarkers for patient stratification and treatment selection. Standard biomarkers often pertain to costly and slow genetic tests. However, recent work has shown that relevant biomarkers can be extracted from these images using convolutional neural networks (CNNs). The CNN-based biomarkers predicted colorectal cancer patient outcomes comparably to gold standards. Extracting CNN-biomarkers is fast, automatic, and of minimal cost. CNN-based biomarkers rely on the ability of CNNs to recognize distinct tissue types from microscope whole slide images. The quality of these biomarkers (coined 'Deep Stroma') depends on the accuracy of CNNs in decomposing all relevant tissue classes. Improving tissue decomposition accuracy is essential for improving the prognostic potential of CNN-biomarkers. In this study, we implemented a novel training strategy to refine an established CNN model, which then surpassed all previous solutions . We obtained a 95.6% average accuracy in the external test set and 99.5% in the internal test set. Our approach reduced errors in biomarker-relevant classes, such as Lymphocytes, and was the first to include interpretability methods. These methods were used to better apprehend our model's limitations and capabilities.

The effect of neural network architecture on virtual H&E staining: Systematic assessment of histological feasibility

Conventional histopathology has relied on chemical staining for over a century. The staining process makes tissue sections visible to the human eye through a tedious and labor-intensive procedure that alters the tissue irreversibly, preventing repeated use of the sample. Deep learning-based virtual staining can potentially alleviate these shortcomings. Here, we used standard brightfield microscopy on unstained tissue sections and studied the impact of increased network capacity on the resulting virtually stained H&E images. Using the generative adversarial neural network model pix2pix as a baseline, we observed that replacing simple convolutions with dense convolution units increased the structural similarity score, peak signal-to-noise ratio, and nuclei reproduction accuracy. We also demonstrated highly accurate reproduction of histology, especially with increased network capacity, and demonstrated applicability to several tissues. We show that network architecture optimization can improve the image translation accuracy of virtual H&E staining, highlighting the potential of virtual staining in streamlining histopathological analysis.