Automatic spine exposure segmentation in stereovision imaging using SAM 2 for driving image updating pipeline for preoperative to intraoperative registration in open spine procedures

<strong>Background</strong>: Surgical navigation is growing to become the standard of care in spine surgery. However, spinal motion occurring from either preoperative-to-intraoperative patient positioning or from movement after intraoperative registration can degrade the navigation accuracy. A level-wise, stereovision-based image updating pipeline has been developed to account for this motion registering each spinal level of preoperative imaging to the intraoperative stereovision surface. <strong>Objective</strong>: We aim to automate the detection of the spine exposure in a series of stereovision images captured during an open spine procedure in order to remove current user-based, manual methods. These images serve as input to the imageupdating pipeline. The Segment Anything Model 2 (SAM2) is leveraged for the automation of this detection. <strong>Methods</strong>: We obtained preoperative CT, intraoperative CT, and tracked stereovision images of a cadaver swine’s spine. SAM2 automatically segments the exposure of the spine in an initial stereovision image and propagates the exposure mask across subsequent images. Information from depth maps are used to refine the masks further creating an automatic mask generation scheme that informs the input to a level-wise registration pipeline. Results: The automated masks perform comparably to a manually segmented mask for level-wise registration accuracy. For levels with sufficient exposure and correspondence, the SAM2 based masks and the manual segmentation masks drive the pipeline to a 2.06 mm +/- 0.49 mm and 1.98 mm +/- 0.95 mm registration accuracy respectively. <strong>Conclusions</strong>: The automated method of spine exposure detection in stereovision aids to remove user dependance from the input to our level-wise registration pipeline in open spine surgery. SAM2 provides good zero-shot inference for stereovision images of the exposed spine surgical site.