A semiautomatic method to identify vertebral end plate lesions (Schmorl's nodes)

Background contextThere are differences in the definitions of end plate lesions (EPLs), often referred to as Schmorl's nodes, that may, to some extent, account for the large range of reported prevalence (3.8%–76%).PurposeThe purpose of this study was to develop a technique to measure the size, prevalence, and location of EPLs in a consistent manner.Study design/settingThis study proposed a method using a detection algorithm that was applied to five adolescent females (average age, 15.1 [range, 13.0–19.2] years) with idiopathic scoliosis (average major Cobb angle, 60° [range, 55°–67°]).MethodsExisting low-dose, computed tomography scans were segmented semiautomatically to extract three-dimensional morphology of each vertebral end plate. Any remaining attachments to the posterior elements of adjacent vertebrae or end plates were then manually sectioned. An automatic algorithm was used to determine the presence and position of EPLs.ResultsEnd plate lesions were identified in 15 of the 170 (8.8%) end plates analyzed with an average depth of 3.1 mm. Eleven of the 15 EPLs were seen in the lumbar spine. The algorithm was found to be most sensitive to changes in the minimum EPL gradient at the edges of the EPL.ConclusionsThis study describes an imaging analysis technique for consistent measurement of the prevalence, location, and size of EPLs. The technique can be used to analyze large populations without observer errors in EPL definitions.