Opportunities for deep learning techniques to advance the histological analysis of preclinical models of osteoarthritis beyond ordinal rank systems

Abstract Tools for assessing disease progression are needed to identify and confirm new mechanisms driving osteoarthritis (OA) progression and guide therapeutic development. This review focuses on the advantages and feasibility of leveraging deep learning techniques for quantitative analysis of rodent knee histology in preclinical OA models. Previous studies using radiographs and micro-CT have used deep-learning driven analysis, including convolutional neural networks (CNNs), to evaluate disease progression in OA patients and preclinical models. However, the use of these tools to analyze histology in rodent OA models is limited. The discrepancy between clinical and preclinical histological quantification partly relates to the size constraints of imaging, where the larger scale of clinical samples requires complex mapping and multiple samples to obtain a complete picture of the entire joint. Rodent OA samples provide a view of the entire joint on an individual slide and thereby enable the quantification of changes across the width of the joint. Here, we discuss approaches for using CNN-based pipelines to quantify and visualize joint remodeling in rodent OA models, complementing existing grading schemes and potentially providing insight into mechanisms driving joint pain and disability.