Deep learning-based segmentation of mineralized cartilage and bone in high-resolution micro-CT images

Léger, Jean [UCL] Leyssens, Lisa [UCL] De Vleeschouwer, Christophe [UCL] Kerckhofs, Greet [UCL]

High-resolution 3D micro-CT imaging is a powerful tool for the visualization of the mineralized tissues. However, it remains challenging to discriminate automatically between mineralized cartilage and bone as they have similar greyscale values. Currently, manual contouring is still the standard way to segment these two tissues but it is time-consuming and user-biased. In this work, we have optimized a 3D fully convolutional neural network, i.e. U-net, to automatically segment mineralized cartilage from bone in high-resolution micro-CT images of the Achilles tendon-to-bone interface. Using the 3D U-net, we reach an average Dice Similarity Coefficient of 0.85 compared to manual annotations for twelve 3D datasets. The proposed method shows comparable results to a 2D U-net approach while ensuring better 3D segmentation consistency. We also found that reducing the resolution of the 3D micro-CT images for the network training did not importantly impact the performance while considerably reducing the training time.