Advancing Accuracy in Orthopedic Imaging

AI-Driven Segmentation on Weight Bearing CT

Introduction

Accurate segmentation is critical for orthopedic workflows, including preoperative planning, patient-specific instrumentation, and 3D modeling. This internal investogation by CurveBeam AI evaluates the accuracy of AI-driven segmentation compared to manual annotation on CBCT imaging and assesses whether higher-dose imaging provides a meaningful advantage.

Key Findings

Automated segmentation demonstrated sub-millimeter accuracy compared to manual annotation across all specimens.

1. Mean surface differences were approximately 0.15–0.29 mm depending on protocol
2. Results were consistent across multiple anatomical structures
3. Accuracy remained within accepted tolerances for orthopedic applications

As demonstrated in the study,

“When applied to weight bearing CT imaging, Atlas behaves like a highly consistent ‘second expert,’ with average discrepancies well below 0.5 mm.”

Protocol Comparison

The investigation evaluated segmentation across standard-dose and higher-dose CBCT protocols.

1. Mean differences between protocols remained close to zero
2. No consistent directional bias was observed
3. Variability remained narrow across most specimens

These findings indicate that standard-dose imaging produces segmentation outputs that are effectively interchangeable with higher-dose imaging in most cases.

Performance in Complex Cases

In the most anatomically complex specimen:

1. Standard-dose mean difference was approximately 0.80 mm
2. Higher-dose mean difference improved to approximately 0.20 mm
3. Variability was reduced with higher-dose imaging

This suggests that higher-dose protocols may provide improved agreement with manual segmentation in complex anatomy.

Click here to download the full whitepaper.