CT Images-Based Automatic Path Planning for Pedicle Screw Placement Incorporating Anatomical and Biomechanical Considerations

Optimal planning of the pedicle screw path is a key guarantee of safe operation during the pedicle screw placement surgery. Existing semi-automatic or automatic path planning methods are mainly based on anatomical information of the vertebrae rather than mechanical information. If the biomechanical information that traditionally obtained through finite element analysis (FEA) could be incorporated into the automatic path planning procedure, it is promising to increase the clinical rationality of the planned surgical path and to reduce the risk of potential surgical failure. FEA is complex and computationally time-consuming, so it is difficult to directly apply the biomechanical information obtained from FEA to path planning. As the neural networks have nonlinear mapping capability to compute and predict complex features, this paper designs a two-stage automatic path planning method incorporating CT anatomical information with biomechanical information. In the first stage, the pre-processed CT images with the stress values obtained from the FEA are fed into the mechanical distributions generating subnetwork (MDGS) for training. In the second stage, CT images, the output of the MDGS and path (entry and directional points) are fed into the path points location subnetwork (PPLS). Finally, the initial paths that obtained from the PPLS are post-processed by the entry point relocating module (EPRM). The normalized mean absolute error (NMAE) of the stress values is 0.555%. The average mean squared error (MSE) of the relocated entry points and directional points are 2.61 and 2.851 voxels, respectively. The results demonstrate the effectiveness of the planned surgical path.