Laboratory-based osteoarthritis (OA) research usually includes study of cartilage degeneration that happens at the early disease stage. Microscopic-level imaging methods are commonly used to study morphology and composition of articular cartilage and specific features of chondrocytes. However, objectively studying those cartilage and chondrocyte features has been limited by lack of automated image analysis algorithms able to accurately analyze microscopic images, and when analyzed manually, the techniques applied are usually laborious and prone to error and bias.

This thesis work focused on developing, validating, and implementing different image analysis algorithms for three microscopic-level imaging methods commonly used to study the degeneration of cartilage. First, an algorithm was developed to segment cartilage from contrastenhanced micro-CT images to facilitate study of cartilage morphology and composition. Second, an image analysis framework was developed to segment chondrocytes from fluorescence confocal microscopic images to study live chondrocyte density and function. Third, a deep-learning-based method was developed to segment chondrocytes and identify chondrocyte clones from cartilage histology images to study disease-related changes in cartilage cellularity.

All of the algorithms developed were found to be fast and validated to be accurate when compared to the manual analysis versions of the work as performed by expert researchers. When these algorithms were applied in OA studies, they were able to characterize differences in tissue morphology and chondrocyte function between injured and non-injured cartilage, and document progressive cellularity changes associated with long-term disease progression. The accurate quantitative analysis from the algorithms developed here benefit OA research by providing a less biased assessment of microscopic-level disease changes.