Lumbar spine localisation method based on feature fusion

To eliminate unnecessary background information, such as soft tissues in original CT images and the adverse impact of similarity adjacent spines on lumbar image segmentation surgical path planning, a two-stage approach for localising segments is proposed. First, based multi-scale feature fusion technology, non-linear regression method used to achieve accurate localisation overall spatial region spine, effectively eliminating useless tissues. In second stage, we directly realised precise positioning each segment spine space method, thus interference caused by spine. The 3D Intersection over Union (3D_IOU) main evaluation indicator accuracy. On an open dataset, 3D_IOU values 0.8339 ± 0.0990 0.8559 0.0332 first stages, respectively achieved. addition, average time required proposed two stages 0.3274 0.2105 s respectively. Therefore, performs very well terms both precision speed can improve accuracy effect planning.