A Deep-Learning Method Using Intraoperative X-ray Images for Locating Distal Holes of Intramedullary Nails

Based on two intraoperative X-ray images at different angles, a method to accurately locate the distal hole is proposed through deep feature extraction and deep regression of the hole axis. Firstly, the nail’s contour is extracted by the object detection algorithm, and the deep neural network is used to predict the projection of the hole’s axis in the imaging plane. Then, the 3D axis pose of the distal hole is preliminarily determined according to the dual-plane intersection. Finally, using the contour information of the nail, the covariance matrix adaptation evolutionary strategies algorithm is used for pose iterative correction. The experiments are carried out in the simulated and clinical environments, and the distal hole’s axis calculated by this method is compared with the actual hole’s axis. In the simulated environment, the average distance error is 0.34 mm, and the average angle error is 0.35 °. Furthermore, the clinical experimental results show that the average distance error is 0.68 mm, and the average angle error is 0.72 °. The method can meet the actual surgical needs of distal hole location; and improves the efficiency of location and planning in the distal locking nail surgery.