Segmentation of the lacunar canalicular network in diabetic rat bone

The lacunar canalicular network (LCN) is a 3D microscopic structure in bone consisting of various features essential to maintaining bone health. This network may be disrupted in diseases that impact the bone, such as diabetes. The LCN can be imaged using confocal laser scanning microscopy. With these images, proper segmentation is required to conduct quantitative analysis to detect whether there is a disruption in the lacunar canalicular network in rats with diabetes compared to rats without diabetes. Segmentation of the LCN is a challenging task due to noise and non-uniform brightness in the image. To overcome these challenges, we identified a combination of image filters to accelerate the segmentation and improve the accuracy of segmentation. To further accelerate the segmentation process, we explored deep learning as a solution to automatically segment the images. Good segmentation was achieved using the U-net neural network architecture. The U-net segmented images with minor manual adjustments were of acceptable quality for further analysis.