Evaluation of a semi-automated software program for the identification of vertebral fractures in children

Abstract

To assess observer reliability and diagnostic accuracy in children, of a semi-automated six-point technique developed for vertebral fracture (VF) diagnosis in adults, which records percentage loss of vertebral body height. Using a semi-automated software program, five observers independently assessed T4 to L4 from the lateral spine radiographs of 137 children and adolescents for VF. A previous consensus read by three paediatric radiologists using a simplified algorithm-based qualitative technique (i.e., no software involved) served as the reference standard. Of a total of 1,781 vertebrae, 1,187 (67%) were adequately visualised according to three or more observers. Interobserver agreement in vertebral readability for each vertebral level for five observers ranged from 0.05 to 0.47 (95% CI: -0.19, 0.76). Intra-observer agreement using the intraclass correlation coefficient (ICC) ranged from 0.25 to 0.61. The overall sensitivity and specificity were 18% (95% CI: 14-22) and 97% (95% CI: 97-98), respectively. In contrast to adults, the six-point technique assessing anterior, middle, and posterior vertebral height ratios is neither satisfactorily reliable nor sensitive for VF diagnosis in children. Training of the software on paediatric images is required in order to develop a paediatric standard that incorporates not only specific vertebral body height ratios but also the age-related physiological changes in vertebral shape that occur throughout childhood.