Simplified, ‘single-step’ segmentation and quantification of bone marrow oedema using deep learning

We present a deep learning workflow which segments and quantifies bone marrow oedema (BMO) in a single deep learning step. Detection and quantification of BMO plays a crucial role in diagnosis and monitoring of patients with inflammatory diseases of the skeleton, including spondyloarthritis (SpA), and various segmentation methods have been developed to facilitate BMO quantification. However, previous attempts have used multiple algorithms in sequence to achieve satisfactory performance. To improve on this, we propose a simplified approach that avoids the need for such sequential algorithms, thus simplifying the workflow, eliminating potential error sources and reducing the need for human input.