Development and validation of an artificial intelligence model to accurately predict spinopelvic parameters

Area of Science:

• Spinal imaging analysis
• Artificial intelligence in medicine
• Radiographic parameter measurement

Background:

• Spinopelvic alignment is crucial for clinical outcomes in spinal conditions.
• Manual measurement of spinopelvic parameters is time-consuming and prone to interobserver variability.
• Existing automated tools often require manual user input, limiting efficiency.

Purpose of the Study:

• To introduce SpinePose, a novel AI tool for automatic spinopelvic parameter prediction.
• To evaluate the accuracy and reliability of SpinePose in measuring key spinopelvic parameters.
• To demonstrate the potential of AI in streamlining spinal imaging analysis.

Main Methods:

• SpinePose was trained on 761 sagittal whole-spine radiographs.
• The AI model predicted 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).
• Accuracy was assessed by comparing SpinePose predictions to expert reviewers on a separate test set of 40 radiographs, using median errors and intraclass correlation coefficients (ICCs).

Main Results:

• SpinePose demonstrated high accuracy with low median errors for all measured parameters (e.g., SVA 2.2 mm, PT 1.3°, PI 2.2°).
• The AI tool exhibited excellent interrater reliability, with ICCs ranging from 0.91 to 1.0.
• Performance was comparable to that of fellowship-trained spine surgeons and neuroradiologists.

Conclusions:

• SpinePose accurately and reliably predicts spinopelvic parameters from spinal radiographs.
• The AI tool eliminates the need for manual user entry, enhancing efficiency.
• SpinePose has the potential to significantly improve patient selection and surgical planning in spinal care.