An automated pipeline for digital phenotyping of brightfield bone biopsy images to generate feature maps for static histomorphometry.
Diagnostic bone biopsies are essential for evaluating metabolic bone disorders such as osteoporosis, osteomalacia, hyperparathyroid bone disease, and renal osteodystrophy. While CT and MRI provide non-invasive, 3D information on bone structure, only histopathology reliably shows bone formation and resorption. Undecalcified bone histology with fluorochrome labeling remains the gold standard for evaluating bone turnover, volume, and mineralization. Traditionally, quantification relies on manual annotation and tracing of relevant tissue structures by experienced pathologists. This process is time-consuming and prone to subjectivity and operator variability. Existing automated or semi-automated quantification methods are still limited and require some manual work and processing time.
Emory inventors have developed the first automated deep learning-based pipeline to trace tissue and cellular structures and extract texture patterns embedded in images of undecalcified bone samples, known as ADAM. The pipeline rapidly generates delineated tissue and cell maps for up to 20 images in less than a minute. The automated tracings allow quicker and easier processing of bone biopsy images for diagnostic reporting. With further evaluation, ADAM could be integrated into existing clinical routines to enhance pathology workflows and contribute to improved diagnostic insights into bone biopsy evaluation and reporting. Publication Bharadwaj. S. et al. (2025). Journal of Bone and Mineral Research Plus, 9(4). https://doi.org/10.1093/jbmrpl/ziaf028