Purpose
When extracting histomorphometric parameters of trabecular bone from in vivo images,
a grayscale approach is advantageous,
since the estimations are not affected by segmentation errors.
In this study,
we compare trabecular thickness (Tb.Th) computed in binary vs.
grayscale from patients imaged in vivo.
Methods and Materials
Legs from 8 patients were imaged by High Resolution peripheral Quantitative Computer Tomography (HRpQCT) with a 3D isotropic resolution of 82µm.
A region of interest was selected from the tibia.
Cortical bone was segmented as proposed in [1] and trabecular bone was segmented using adaptive region growing (ARG) [2] and Otsu’s threshold.
Local Tb.Th was computed using sphere-fitting [3] and granulometry [4] for the segmented and grayscale images,
respectively.
In order to assess local differences among methods,
we compared the mean of Tb.Th at smaller...
Results
The mean (std) of trabecular thickness for the bone samples are 0.337(0.095),
0.462(0.124) and 0.426(0.085) millimeters for Otsu,
ARG and grayscale,
respectively.
Table 1 presents the mean correlations of the linear regression between the methods.
As shown in the table,
estimations of Tb.Th are very sensitive to the segmentation algorithm with Otsu and ARG being poorly correlated at a smaller scale.
On the other hand,
strong correlations can be observed between the grayscale method and both binary methods for all samples at different scales (cube...
Conclusion
Computing trabecular thickness in human tibiae at different scales,
we found close correlations between gray-scale and binary methods.
Estimations in binary are sensitive to the segmentation method.
Direct estimation in grayscale avoids this step,
potentially making them more suitable for images acquired in vivo.
References
[1]Chowdhury M,
Jörgens D,
Wang C,
Smedby Ö and Moreno R.
Segmentation of cortical bone using fast level sets.
Proc.
SPIE Medical Imaging 2017,
1013327
[2] Revol-Muller C,
Peyrin F,
Carrillon Y,
et al.
Automated 3D region growing algorithm based on an assessment function.
Pattern Recognition Letters,
2002,
23(1).
[3] Hildebrand T,
Rüegsegger P.
A new method for the model‐independent assessment of thickness in three‐dimensional images.
Journal of microscopy 185 (1),
67-75
[4]Moreno R,
Borga M,
Smedby Ö.
Estimation of trabecular thickness in gray-scale images...
Personal Information
Michael Platten (1,2),
Manish Chowdhury (1),
Örjan Smedby (1,2,3) & Rodrigo Moreno (1)
1: School of Technology and Health at KTH Royal Insitute of Technology,
SE; 2: Karolinska Institutet,
SE; 3: Linköping University,
SE