Cell quantification at the osteochondral interface from synchrotron radiation phase contrast micro-computed tomography images using a deep learning approach

Abstract

Abstract Osteochondral interface consists of two tissues: the calcified cartilage (CC) containing chondrocytes, and subchondral bone (SCB) containing osteocytes that interact with each other. In this study, we propose a new method for the three-dimensional (3D) segmentation of chondrocyte and osteocyte lacunae in CC and SCB from human knees, imaged using high resolution (650 nm) synchrotron radiation phase contrast micro-computed tomography (SR phase contrast micro-CT). Our approach is based on marker-controlled watershed (MCW) algorithm combined with a deep learning method (nnU-Net). We demonstrate that incorporating nnU-Net into the MCW process improves the identification and segmentation of cell lacunae. Using this method, we analyzed a subsample of fifteen cores extracted from the central area of the medial tibial plateaus. Several quantitative parameters (lacunar volume fraction, number density, volume, anisotropy and structure model index of cell lacunae) were measured to compare 10 control and 5 osteoarthritic knees. While no significant differences were observed in chondrocytes, osteocytes showed lower anisotropy (width/depth) and a tendency toward more spherical shapes in the osteoarthritic group compared to the control group. The phase contrast underlying the chondro-osseous border allowed to analyze separately CC from SCB in SR phase contrast micro-CT images. This new method may help to better understand the cellular behavior at the osteochondral interface in osteoarthritis.

Keywords

• Synchrotron radiation phase contrast micro-computed tomography
• Watershed
• Deep learning
• Chondrocytes and osteocytes
• Segmentation
• Osteoarthritis