Augmented Reality Spatial Registration for Spine Surgery: A 2D/3D Image-Based Approach

Results\nThis article establishes a successful registration threshold of 8 mm, indicating that a final target registration error (mTRE) of less than 8 mm constitutes successful registration. The spine model registration achieved a success rate of 0.9. Both patients' medical image registrations were successful. The image registration experiment is illustrated in Figure 11. Group a depicts the medical image registration experiment for the spinal model, while groups b and c represent the medical image registration experiments for the two patients. Figure 1 presents the X-ray image, Figure 2 displays the generated DRR image, and Figure 3 shows the image after the DRR image is registered with the X-ray image. Figure 12 presents the results of image registration accuracy.\n\nFigure 11: Image registration experiment\n(a) Medical images of the model. (b) Medical images of patient number one. (c) Medical images of patient number two. (1) shows the X-ray image, (2) shows the generated DRR image, and (3) shows the image after registration.\n\nFigure 12: mTRE results of image registration\nIn the model experiment, the box plot of surface registration error is displayed in Figure 13, with 10 groups of experiments conducted. The results indicate that the surface registration error falls within the range of 0.361 to 0.612 millimeters, and the overall average surface registration error is 0.501 millimeters.\n\nFigure 13: Surface registration error box plot\n\n4. Discussion\nThis study proposes a spatial registration method based on 2D/3D registration to achieve spatial registration of the medical image coordinate system and the patient coordinate system. This registration method aims to provide accurate and reliable image guidance for complex spine surgeries. In the 2D/3D image registration experiment, the success rate of spine model registration was 0.9, and the medical image registrations of both patients were successful. In the spinal model verification experiment, the average surface registration error was 0.501 mm. The accuracy of pedicle screw positioning in traditional surgery is 3.89 mm [30], while the experimental measurement accuracy is higher.\n\nCurrently, the primary registration methods employed in surgical navigation systems are point registration based on physical anatomical features and marker-based registration. Point registration based on physical anatomical features necessitates invasive exposure of the target anatomy and extensive contact with it. The patient is rigidly connected to the marker, and preoperative CT is acquired together. This registration method can cause additional trauma to the patient, which contradicts the goal of minimally invasive surgery. The practice of attaching markers to the skin surface can lead to errors due to displacement between the skin and the target anatomy during surgery. These markers are typically affixed to the patient's skin prior to surgery, and their positions may shift between imaging and actual registration. Moreover, the arrangement of reference points on the skin surface will also influence registration accuracy, and even with optimized spatial placement methods, TRE within a range of 3 mm may still occur [31].\n\nCompared to these two methods, the image-based 2D/3D registration spatial registration method presented in this paper (utilizing intraoperative X-ray images to resolve registration) offers several advantages. Firstly, there is no need for the marker to come into direct contact with the patient's anatomical area, avoiding additional trauma to the patient. Secondly, since the image-based approach only requires the acquisition of a few X-ray images, surgical time is reduced. Thirdly, there is no need for manual sketching and segmentation of the marker location, reducing the complexity of the procedure and minimizing dependence on the doctor's experience.\n\nFurthermore, there are still some issues worth considering and discussing. In the 2D/3D registration process, some technical obstacles hinder the practical application of registration, such as the uncertainty of external/internal calibration of X-ray imaging devices, which can affect the results of spatial registration. In the spatial registration process of this study, a personalized marker was used, and the coordinate values of its four corner points were utilized to solve the transformation matrix. Employing three or more markers and considering the coordinate values of each marker as a point can further enhance registration accuracy. The display device for augmented reality in this study is a movable screen, but doctors still need to shift their line of sight between the screen and the patient. Developing a surgical navigation system based on HMD display can improve the immersion and portability of the doctor's surgery.\n\nConclusions\nThe augmented reality spatial registration method proposed in this study, based on 2D/3D registration technology, can be implemented in an augmented reality-based spinal surgery navigation system for registering medical images and patients. This intraoperative method of obtaining the relationship between the medical image and the patient avoids the trauma caused by rigid connections before surgery. There is also no need for manual removal of the point, preventing additional infection risk to the patient and reducing the operation time. During the operation, the position of the virtual model in the world coordinate system can be dynamically registered by simply allowing the monocular camera to view the personalized markers, and the spinal model can be registered with the reconstructed 3D virtual image, minimizing interference with the current surgical workflow. The key technologies established demonstrate significant innovation. In future research, specific characteristics and requirements of clinical practice will be targeted to promote the technology's advancement and application in the clinical setting.