Generalizing convolution neural networks on stain color heterogeneous data for computational pathology

Integrating artificial intelligence in bedside care for covid-19 and future pandemics

“…• Variations in healthcare institution policies, resources, expertise, patient populations and differences between prepandemic and postpandemic operations require careful consideration in generalising AI solutions 7 These challenges are magnified across geographical areas, where concepts might need to be translated between languages, and adjustments made for baseline patient characteristics and technological capabilities. In some instances, differences in technological capabilities could prevent low income countries from benefiting from AI, or require specialised approaches towards knowledge transfer.…”

stainlib: a python library for augmentation and normalization of histopathology H&E images

“…The second approach refers to “stain” or “color augmentation methods”, which create new synthetic samples to increase the training dataset size, creating more robust models regarding color variations. There are novel image processing and machine learning techniques reported in the literature to deal with color heterogeneity, improving classification, and segmentation performance for various tissue types [19, 6, 14, 4, 11]. While the specific normalization technique depends on the task to solve [19, 4], recent work has reported consistent improvements in performance and robustness to external datasets employing color augmentation techniques [19, 4] or a combination of normalization and augmentation [14].…”

“…It is now well known that deep learning classification and segmentation models for histopathology yield better results when data augmentation is used [19, 6, 14, 7]. The benefits of data augmentation might be intuitive in training deep learning models, where the larger the amount of data the model is fed with, the more variations the model is exposed to.…”

“…Section 3 presents a qualitative evaluation of the normalization and augmentation methods included in . Quantitative evaluation of these methods has also been performed in previous research from the authors [14, 6, 19].…”

Evaluating Registrations of Serial Sections With Distortions of the Ground Truths

“…A convolutional neural network, transforming "clean" images into damaged ones with some kind of a style transfer [156,157,158,159] is an interesting idea. We sought for a functional and well-defined image deformation that allowed for using transformed images as a benchmark input.…”