Interrater Reliability of Partially Automated Segmentation of Spinal Radiographs in Adult Spinal Deformity Patients

Spinopelvic radiographic measurements are fundamental in the evaluation and treatment of adult spinal deformity (ASD). Measuring spinopelvic parameters has become quicker and more reliable with the development of automated software, such as the Knowledge Environment for Orthopedic Personalized Surgery (KEOPS), but reliability in clinical practice may be limited by variable image quality. The aim of this study is to determine the interrater reliability of partially automated measurement of spinopelvic parameters for patients with ASD using the KEOPS software.

Representative preoperative and postoperative full-length spinal radiographs of 5 patients who underwent surgical correction of ASD were selected from an imaging database. Five fellowship-trained spine surgeons, 1 spine fellow, and 1 orthopedic surgery resident performed manual segmentation on each image, followed by automated measurement of spinopelvic parameters by the KEOPS software. Interrater reliability was evaluated using intraclass correlation coefficients (ICCs) and 95% confidence intervals (CIs). The proportion of images for which segmentation was successfully completed by the KEOPS software was analyzed.

The ICCs indicated excellent interrater reliability for all spinopelvic parameters. The highest ICC was observed for pelvic tilt (0.994; 95% CI, 0.985-0.998), and the lowest was observed for C2-7 lordosis (0.954; 95% CI, 0.851-0.995). Segmentation completion rates decreased in the cervical spine, ranging from 88%-97% in the sagittal plane and 51%-61% in the coronal plane because of poor visualization.

The partially automated system using the KEOPS software demonstrates excellent interrater reliability for calculating spinopelvic parameters in patients with ASD. Segmentation involving the cervical spine remains less consistent on both sagittal and coronal radiographs.