(Semi-)Automatic Segmentation of Vertebral Bone Marrow Lesions on Magnetic Resonance and Dual Energy Computed Tomography Images: a Literature Review

Abstract

Bone marrow lesion syndrome (BMLS) describes the phenomenon of strong signal intensity changes in the bone marrow in fluid-sensitive magnetic resonance imaging (MRI). BMLs have taken up a central presence in many different diseases affecting the musculoskeletal system, including the spine. As the condition of the spine is of strong influence to the overall body’s health, detection and characterization of BMLs could help in preventing the progression of spine related diseases. With the increased clinical availability of dual energy computed tomography (DECT), an alternative method for imaging of BMLS has presented itself through virtual non-calcium (VNCa) techniques. In this paper we explore the current state of both imaging modalities for (semi-)automatic segmentation of BMLs in the spine, as well as a variety of approaches to the segmentation of individual vertebrae that could be used to develop a fully automated process for BML segmentation. We found that for both MRI and DECT, automatic BML segmentation is still in its infancy, with few studies attempting the task. Additionally, manual segmentations of BMLs, which act as the ground truth for evaluating automated approaches, generally show significant inter- and intraobserver variability. However, with CT being the modality of choice in the traumatic setting, its potential for further developments in material decomposition and various other advantages that DECT poses over MRI, DECT could become the future modality of choice in both qualitative and quantitative imaging of BMLS.