Background High-Resolution peripheral QCT (HR-pQCT) imaging has a higher sensitivity in the detection of breaks compared to conventional radiography (1). Cortical bone in the finger joints is very thin ($≈ $100μm) and it is therefore possible that the HR-pQCT is not able to detect these thin structures. An automatic algorithm that is based on binary images can therefore falsely identify these regions as breaks. To investigate the extent of this error, cortical break detection on HR-pQCT was compared to that on μCT with a higher resolution.
Objectives To investigate the proportion of falsely detected breaks with a semi-automatic algorithm on HR-pQCT compared to μCT.
Methods Nineteen finger joints of ten human female cadaveric index fingers (mean age ± SD; 85.1 ± 9.6 years) with unknown medical history were imaged by HR-pQCT and μCT (82 and 18μm isotropic voxel sizes, respectively). A semi-automatic algorithm was applied to HR-pQCT and μCT for the detection of cortical breaks. First, the outer margin of the bone structure was contoured. Second, the bone within 0.25mm from this contour was selected as cortical region. Last, different sizes for defining a cortical break (>0.50mm, >0.66mm and >0.82mm) were applied and evaluated. μCT images were registered to HR-pQCT in order to compare the locations of the detected breaks. The false discovery rate (FDR) of breaks detected on HR-pQCT was calculated, with μCT as reference.
Results The number of breaks depended on the image modality and chosen break size, and varied between 0.8 and 20.8 breaks per joint (table 1).
The locations of the detected breaks on HR-pQCT and μCT generally correspond well (Fig 1-I). On HR-pQCT, however, breaks may be detected as large breaks, whereas on μCT these large breaks appear to represent a combination of several small breaks (Fig 1-II). With larger break diameters as a cut-off for both techniques, this resulted in the detection of breaks on HR-pQCT which were not detected on μCT. Therefore, the FDR for the detection of a break of the same size was sufficient for small breaks (37.9%), but poor for large break sizes (86.7%). However, when comparing a larger break on HR-pQCT with the smallest minimal break size on μCT it was found that only 4.4% of the detected breaks on HR-pQCT were falsely detected (table 1).
Conclusions Because of its limited resolution, a single break region detected on HR-pQCT can appear as several smaller breaks on μCT. When accounting for this effect, however, excellent agreement is found between cortical break detected with the algorithm using HR-pQCT and μCT. We thus conclude that the use of HR-pQCT in combination with our semi-automatic algorithm is a promising tool for early detection and monitoring of the number of small cortical breaks in finger joints.
Stach CM, A&R.2010 Feb; 62(2):330–339
Disclosure of Interest M. Peters: None declared, A. Scharmga: None declared, A. van Tubergen: None declared, B. van Rietbergen Consultant for: Scanco Medical AG, R. Weyers: None declared, D. Loeffen: None declared, J. Van den Bergh: None declared, P. Geusens: None declared