Automated analysis of rabbit knee calcified cartilage morphology using micro‐computed tomography and deep learning. Issue 2 (29th March 2021)

• Record Type: Journal Article
• Authors: Rytky, Santeri J. O.; Huang, Lingwei; Tanska, Petri; Tiulpin, Aleksei; Panfilov, Egor; Herzog, Walter; Korhonen, Rami K.; Saarakkala, Simo; Finnilä, Mikko A. J.
• Is Part Of: Journal of anatomy. Volume 239:Issue 2(2021)
• Page Start: 251
• Page End: 263
• Publication Date: 2021-03-29
• Subjects: animal models -- bone -- histology -- osteoarthritis -- segmentation; Anatomy -- Periodicals; 571.3
• DOI: 10.1111/joa.13435
• Languages: English
• ISSNs: 0021-8782

Abstract

Abstract: Structural dynamics of calcified cartilage (CC) are poorly understood. Conventionally, CC structure is analyzed using histological sections. Micro‐computed tomography (µCT) allows for three‐dimensional (3D) imaging of mineralized tissues; however, the segmentation between bone and mineralized cartilage is challenging. Here, we present state‐of‐the‐art deep learning segmentation for µCT images to assess 3D CC morphology. The sample includes 16 knees from 12 New Zealand White rabbits dissected into osteochondral samples from six anatomical regions: lateral and medial femoral condyles, lateral and medial tibial plateaus, femoral groove, and patella ( n = 96). The samples were imaged with µCT and processed for conventional histology. Manually segmented CC from the images was used to train segmentation models with different encoder–decoder architectures. The models with the greatest out‐of‐fold evaluation Dice score were selected. CC thickness was compared across 24 regions, co‐registered between the imaging modalities using Pearson correlation and Bland–Altman analyses. Finally, the anatomical CC thickness variation was assessed via a Linear Mixed Model analysis. The best segmentation models yielded average Dice of 0.891 and 0.807 for histology and µCT segmentation, respectively. The correlation between the co‐registered regions was strong ( r = 0.897, bias = 21.9 µm, standard deviation = 21.5 µm). Finally, both methods could separate the CC thickness between the patella, femoral, and tibial regions ( p < 0.001). As a conclusion, the proposed µCT analysis allows for ex vivo 3D assessment of CC morphology. We demonstrated the biomedical relevance of the method by quantifying CC thickness in different anatomical regions with a varying mean thickness. CC was thickest in the patella and thinnest in the tibial plateau. Our method is relatively straightforward to implement into standard µCT analysis pipelines, allowing the analysis of CC morphology. In future research, µCT imaging might be preferable to histology, especially when analyzing dynamic changes in cartilage mineralization. It could also provide further understanding of 3D morphological changes that may occur in mineralized cartilage, such as thickening of the subchondral plate in osteoarthritis and other joint diseases. Abstract : We present a micro‐computed tomography‐based method with deep learning segmentation for analyzing calcified cartilage (CC) thickness. The comparison of regions co‐registered to histology yielded a strong Pearson correlation ( r = 0.90). Both methods were able to separate the CC thickness properties between tibia, femur, and patella.