Hidden Variables in Deep Learning Digital Pathology and Their Potential to Cause Batch Effects: Prediction Model Study

Abstract

Background An increasing number of studies within digital pathology show the potential of artificial intelligence (AI) to diagnose cancer using histological whole slide images, which requires large and diverse data sets. While diversification may result in more generalizable AI-based systems, it can also introduce hidden variables. If neural networks are able to distinguish/learn hidden variables, these variables can introduce batch effects that compromise the accuracy of classification systems. Objective The objective of the study was to analyze the learnability of an exemplary selection of hidden variables (patient age, slide preparation date, slide origin, and scanner type) that are commonly found in whole slide image data sets in digital pathology and could create batch effects. Methods We trained four separate convolutional neural networks (CNNs) to learn four variables using a data set of digitized whole slide melanoma images from five different institutes. For robustness, each CNN training and evaluation run was repeated multiple times, and a variable was only considered learnable if the lower bound of the 95% confidence interval of its mean balanced accuracy was above 50.0%. Results A mean balanced accuracy above 50.0% was achieved for all four tasks, even when considering the lower bound of the 95% confidence interval. Performance between tasks showed wide variation, ranging from 56.1% (slide preparation date) to 100% (slide origin). Conclusions Because all of the analyzed hidden variables are learnable, they have the potential to create batch effects in dermatopathology data sets, which negatively affect AI-based classification systems. Practitioners should be aware of these and similar pitfalls when developing and evaluating such systems and address these and potentially other batch effect variables in their data sets through sufficient data set stratification.

Robustness of convolutional neural networks in recognition of pigmented skin lesions

“…While CNN-based image analysis has advantages over human observation with respect to objective and quantitative feature extraction, an obvious drawback is that in contrast to human experts, CNNs have difficulty distinguishing biologically significant features from insignificant features and artifacts. Depending on the data set that is used for CNN training, spurious and unwanted correlations within the training set can be picked up and hamper generalization [19,20,46]. Moreover, deceptively created input images specifically designed to fool a CNN (adversarial attacks) have been shown to pose a real threat [21].…”

Digital pathology and artificial intelligence in translational medicine and clinical practice

“…Various quality control (QC) techniques can be used to overcome preanalytical issues such as variations in slide preparation, origin, and scanner type. One approach is to train individual models of the same architecture to recognize specific variables 73 . Other approaches, such as combining image metrics in a QC application 138 , or transformation of image patches with synthetically generated artifacts 139 , can be used to train an algorithm to recognize different types of histological artifacts.…”

“…Data used for training need to be accurate and as complete as possible in order to maximize predictability and utility 39 . This can be challenging when histological data are obtained from various laboratories, leading to some variability due to factors such as differences in slide preparation (sectioning, fixation, staining, and mounting) 73 , scoring algorithms 18 , and inherent inter-observer variability 74 . These challenges become more apparent when more complex computational analytics methods are used for multiplexed imaging.…”

Deep learning can predict lymph node status directly from histology in colorectal cancer

“…Moreover, a separate analysis of the subgroup with T1 tumours would have been very interesting but was not feasible because of the small number of T1 cases. As in all studies using a CNN to extract features from histological slides, hidden variables such as specimen fixation, slide preparation date, slide origin or scanner type in the digital images might cause batch effects and influence prediction accuracy [35]. Finally, we used WSIs of the operation specimens.…”