Bo-Net: Deep learning-based model for automatic bone stromal cell segmentation of fluorescence microscopy images

Area of Science:

• * Bone biology and spatial analysis
• * Medical imaging and computational pathology
• * Preclinical research and drug discovery

Background:

• * Bone is a dynamic organ susceptible to diseases like cancer.
• * 3D microscopy enables detailed spatial biology studies at the subcellular level.
• * Manual analysis of these images is time-consuming, variable, and lacks standardization.

Purpose of the Study:

• * To develop an automated solution for analyzing multiparametric fluorescence microscopy images of bone.
• * To overcome limitations of manual analysis in terms of time, standardization, and variability.
• * To enable high-throughput investigation of bone pathophysiology.

Main Methods:

• * Application of deep learning for automatic semantic segmentation of bone cells (osteoblasts, osteoclasts, blood vessels) and mineral components.
• * Development and implementation of a neural network architecture named Bo-Net.
• * Training and validation using a dataset of 21,395 fluorescence microscopy images.

Main Results:

• * Bo-Net achieved high performance and segmentation accuracy comparable to experienced biologists (R=0.81-0.99).
• * Automated analysis time was reduced from days to seconds.
• * The tool demonstrated relevance across various experimental and biological contexts.

Conclusions:

• * A fully automated and reliable tool, Bo-Net, has been developed for bone research.
• * This tool optimizes the analysis of complex multiparametric fluorescence microscopy data.
• * Bo-Net enhances the efficiency and reproducibility of spatial biology studies in bone.