Related Experiment Video
Updated: May 14, 2026
06:45
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Improved segmentation accuracy in high-resolution peripheral quantitative computed tomography scans of carpal bones
Michael T Kuczynski1,2,3, Simone Poncioni4, Sarah Elmahdy1,2,3
1McCaig Institute for Bone and Joint Health, University of Calgary, Calgary, AB, Canada.
JBMR Plus
|May 13, 2026
Summary
Adaptive thresholding (AT) accurately segments carpal bone microarchitecture in HR-pQCT scans, outperforming standard methods for improved bone health assessments. This advanced technique enhances precision in measuring trabecular thickness, separation, and bone volume fraction.
Area of Science:
- Biomedical Engineering
- Radiology
- Orthopedics
Background:
- High-resolution peripheral quantitative computed tomography (HR-pQCT) is crucial for assessing bone microarchitecture.
- Segmentation methods significantly impact the accuracy of bone structure measurements from HR-pQCT scans.
- Global thresholding methods struggle with small or under-mineralized bones, like carpal bones, due to limitations in capturing local intensity variations.
Purpose of the Study:
- To compare the accuracy of adaptive local thresholding (AT) against global thresholding methods (Gaussian and Laplace-Hamming filters) for segmenting carpal bones in HR-pQCT scans.
- To evaluate the performance of AT in measuring key trabecular microarchitecture parameters (Tb.Th, Tb.Sp, Tb.BV/TV) and assess spatial agreement with micro-CT (μCT).
Main Methods:
- HR-pQCT scans of 64 ex vivo human carpal bones were analyzed.
- AT parameters were optimized on 24 carpal bones, and performance was validated on 40 carpal bones.
- Segmentation methods included Gaussian filtering with global thresholding, Laplace-Hamming filtering with global thresholding, and AT.
- Micro-CT served as the gold standard for comparison, with images segmented using Gaussian filtering and Otsu's method.
Main Results:
- The AT method demonstrated the lowest absolute and relative errors and bias for all trabecular parameters compared to global thresholding methods.
- AT reduced mean absolute error by 36% for trabecular thickness (Tb.Th), 14% for trabecular separation (Tb.Sp), and 15% for bone volume fraction (Tb.BV/TV).
- AT achieved superior spatial agreement with micro-CT, indicated by a Dice Similarity Coefficient (DSC) of 0.84, Hausdorff distance 95th percentile (HD95) of 0.061 mm, and average symmetric surface distance (ASSD) of 0.018 mm.
Conclusions:
- Adaptive thresholding (AT) significantly outperforms standard Gaussian and Laplace-Hamming global thresholding methods for segmenting carpal bones in HR-pQCT images.
- AT provides more accurate and spatially precise measurements of trabecular microarchitecture in carpal bones, extending validation to this specific anatomical region.
- These findings support the use of AT for improved characterization of bone health in small and potentially under-mineralized skeletal sites.
