Development of a manufacturer-independent convolutional neural network for the automated identification of vertebral compression fractures in vertebral fracture assessment images using active learning

•GE and Hologic are major manufacturers of DXA scanners capable of acquiring vertebral fracture assessment (VFA) images•Convolutional neural networks (CNNs) can accurately identify vertebral fractures on both GE and Hologic VFAs•Training CNNs simultaneously on Hologic and GE VFAs allows for models that generalize to both scan types•Active learning can reduce the time and cost required to train an effective medical image classifier