Abstract
Introduction: Smartphone-based markerless pose estimation may provide a non-radiographic method for evaluating whole-body alignment in patients with adult spinal deformity (ASD). This study investigated whether selected photographic parameters derived from two markerless pose-estimation frameworks correspond to standard radiographic alignment measures.
Methods: This study included 34 patients with ASD who underwent simultaneous acquisition of frontal and lateral smartphone photographs and standing whole-spine plus full-length lower-limb radiographs. Shoulder balance, sagittal trunk tilt, coronal trunk tilt, knee varus, and knee flexion parameters were derived from MoveNet and Apple Vision Framework outputs. Directly comparable photographic-radiographic pairs were evaluated as the primary analysis. For sagittal trunk alignment, right-side-only sagittal trunk tilt angle using ear-hip landmarks was used for the principal agreement comparison with trunk tilt angle (TTA) because landmarks on the non–camera-facing side were less stable in some lateral photographs, whereas bilateral sagittal formulations were examined in sensitivity analyses. Associations with sagittal vertical axis (SVA) and C7-central sacral vertical line (C7-CSVL) were treated as secondary proxy analyses.
Results: Directly comparable photographic parameters showed favorable correspondence with radiographic measures. The principal sagittal agreement comparison was right-side-only STTA_EH versus TTA (parameters derived from Apple Vision Framework [AVF]: mean absolute error [MAE] 1.36°, r = 0.990; parameters derived from MoveNet [MN]: MAE 1.85°, r = 0.988). Shoulder height angle corresponded well with clavicle angle (AVF: MAE 1.27°, r = 0.905). Lower-limb parameters also showed favorable performance, particularly knee varus angle versus hip-knee-ankle angle. Secondary analyses showed that selected trunk parameters were associated with SVA and C7-CSVL, whereas hip height angle versus pelvic obliquity remained weak. Bilateral sagittal formulations were retained for sensitivity analyses.
Conclusions: Smartphone-based markerless photography showed favorable performance for selected whole-body alignment parameters in patients with ASD and may serve as a complementary non-radiographic assessment method. Directly corresponding angular parameters were the most robust findings, whereas proxy parameters should be interpreted cautiously.
