Medical Student University of Michigan Medical School Ypsilanti, Michigan, United States
Introduction: SpinePose was developed in 2024 as a novel artificial intelligence (AI) tool to automatically predict spinopelvic parameters with high accuracy and without the need for manual entry. Our published results demonstrated excellent performance comparable to a fellowship-trained spine surgeon. To date, there are no studies that externally validate the performance of AI spinopelvic parameter measurement tools on data acquired from other institutions. To assess the generalizability of SpinePose, we report its performance on an external set of heterogeneous whole-spine scoliosis x-rays obtained from an outside institution.
Methods: SpinePose was initially trained/validated on a dataset of 761 sagittal whole-spine scoliosis films from a single institution, with expert-level performance on both whole-spine and lumbosacral x-rays. In this study, the existing SpinePose model was used to inference on a new set of 49 whole-spine x-rays acquired at a tertiary academic hospital located out-of-state. Externally acquired x-rays represented a diverse set of images, incorporating both instrumented and non-instrumented patients, and a wide variety of fusion constructs including complex deformity patients. Predicted measures included Sagittal Vertical Axis (SVA), Pelvic Tilt (PT), Pelvic Incidence (PI), Sacral Slope (SS), Lumbar Lordosis (LL), T1-Pelvic Angle (T1PA), and L1-Pelvic Angle (L1PA). Median errors relative to ground truth manual annotations were calculated to determine the model’s accuracy.
Results: Of the 49 images, 39 (79.6%) contained instrumentation, in comparison to 58% in the original SpinePose test set. SpinePose exhibited the following median(IQR) parameter errors relative to ground-truth: SVA: [2.1mm (8.9mm), p=0.72], PT: [2.6° (5.0°) , p=0.41], SS: [3.4° (11.6°) , p=0.10], PI: [7.3° (11.8°) , p=0.03], LL: 6.3° (13.6°) , p=0.48], T1PA: [1.9° (4.0°) , p=0.55], and L1PA: [1.2° (3.5°), p=0.54].
Conclusion : SpinePose was able to accurately predict spinopelvic parameters on an external validation cohort that was generated completely independently from the images the model was trained and validated on. This highlights the generalizability of SpinePose to be implemented on novel images from other institutions and geographic settings with high accuracy and minimal pre-processing. The implementation of AI tools more broadly will help standardize our ability to both deliver and provide spine care and assist with surgical treatment and management of spine patients.