We propose a semi-automated pipeline to joint kinematics in the TMC joint using dynamic CT. Our joint angle results (Table 1) fall within range of previously reported TMC joint angles for similar thumb movements (1, 21, 27). While much of this pipeline has been automated, SCSs were defined using manually placed ALs. Contrary to our original hypothesis, we found that both inter- and intra-rater repeatability was poor to excellent. However, we found that within rater results exhibited lower variability (Fig. 5, Intra-Rater ICC) compared to between rater results (Fig. 5, Inter-Rater ICC). Inter-rater flexion-extension angles demonstrated lower ICC values when compared to inter-rater abduction-adduction angles. Trapezium morphology has been shown to change with age and disease state (28), which may lead to greater variability between AL placement and explain the decreased inter-rater ICC for the flexion-extension angles. An automated method to define TMC joint SCSs based on joint surface geometry has been previously defined (29), however, it is not yet clear how such automated methods for SCS definition are affected by changes in articular morphology. In a post-hoc qualitative analysis, we found that joints with considerable joint degeneration had poorer agreement between raters compared to joints that appeared healthy. In joints that visually appeared healthy, ICC values were moderate to excellent between and within raters (ICC > 0.5).
Table 1
Mean joint angles across three raters
Specimen
|
Joint Angle Mean [Range] (degrees)
|
Abduction-Adduction
|
Flexion-Extension
|
Axial Rotation
|
1
|
34.2 [32.8, 36.9]
|
-24.3 [-25.8, -21.1]
|
-8.2 [-9.6, -6.8]
|
2
|
43.4 [41.1, 45.7]
|
-17.5 [-19.8, -16.0]
|
-4.4 [-6.5, -2.6]
|
3
|
46.9 [44.3, 48.6]
|
-11.9 [-12.8, -10.3]
|
-10.6 [-12.9, -9.7]
|
4
|
46.3 [43.7, 49.8]
|
-3.1 [-5.2, -1.0]
|
-10.4 [-13.2, -7.9]
|
5
|
33.8 [30.7, 37.8]
|
-13.8 [-15.1, -9.5]
|
-1.9 [-3.3, 0.7]
|
6
|
38.2 [36.5, 41.6]
|
-20.3 [-21.6, -18.2]
|
1.7 [-0.8, 4.6]
|
7
|
40.4 [36.9, 47.8]
|
-18.1 [-24.1, -15.1]
|
-7.5 [-9.6, -4.4]
|
8
|
48.7 [46.6, 53.2]
|
-11.5 [-13.5, -10.0]
|
-1.9 [-4.0, 0.5]
|
9
|
34.4 [20.0, 40.1]
|
-26.2 [-28.4, -20.1]
|
-5.9 [-7.9, -1.5]
|
Our results indicate that manual selection of ALs to compute TMC joint angles can be performed reproducibly, provided that the joints being assessed are not severely degenerated and a single rater performs AL placement. In Bland-Altman analysis, we compared each rater’s resulting joint angles to the mean across all raters. Applying a linear regression model to each rater’s Bland-Altman plot revealed very small slopes across all raters and specimen (Table 2). While approximately half of these slopes were statistically significantly different than zero (p < 0.05), the small magnitude of regression slopes indicates that the difference between each rater and the average was not related to the magnitude of the average angle measurement. Further, we found that there was consistent bias between raters and the mean (Fig. 7), representative of expected differences between rater’s AL placement. Whether differences detected between raters is clinically relevant warrants further investigation.
The image processing required for dynamic CT scans introduces new sources of error that are not prevalent in motion capture systems. When evaluating the image registration accuracy visually, image alignment was excellent. After quantitative analysis of the TREs associated with each image registration, we found voxel or sub-voxel level errors, suggesting that only a small portion of the error in the joint angle results between raters was due to the image registration steps. When comparing SCS orientation between raters in the HR-pQCT and dynamic CT image space, we found little difference in the orientation of SCS axes between image spaces. This suggests that any orientation difference between SCSs defined in the HR-pQCT image space are held constant after transforming all data to the dynamic CT space, further reinforcing our confidence that the resulting difference in joint angles between raters are due to rater biases when defining SCSs, not due to the image registration.
Table 2
Slopes of regression lines plotted for Bland-Altman analysis
Specimen
|
Abduction-Adduction Slope
|
Flexion-Extension Slope
|
Axial Rotation Slope
|
Rater 1
|
Rater 2
|
Rater 3
|
Rater 1
|
Rater 2
|
Rater 3
|
Rater 1
|
Rater 2
|
Rater 3
|
1
|
0.03
|
-0.01
|
-0.02
|
-0.02
|
0.02
|
0.01
|
-0.01
|
-0.02
|
0.01
|
2
|
0.01
|
0.01
|
-0.02
|
-0.01
|
0.02
|
0.01
|
0.00
|
-0.04
|
0.00
|
3
|
-0.03
|
-0.02
|
-0.02
|
0.04
|
0.03
|
0.00
|
-0.02
|
-0.01
|
-0.01
|
4
|
0.00
|
0.05
|
0.01
|
0.00
|
-0.05
|
0.04
|
-0.01
|
-0.01
|
-0.05
|
5
|
0.00
|
-0.01
|
0.01
|
0.00
|
0.02
|
-0.04
|
-0.01
|
-0.01
|
-0.01
|
6
|
0.02
|
-0.01
|
0.02
|
-0.01
|
0.00
|
-0.01
|
-0.01
|
0.01
|
0.00
|
7
|
-0.01
|
0.00
|
-0.02
|
0.01
|
-0.01
|
-0.01
|
0.00
|
0.01
|
0.00
|
8
|
0.02
|
-0.01
|
-0.02
|
-0.04
|
0.00
|
0.01
|
0.01
|
0.00
|
-0.02
|
9
|
0.06
|
-0.01
|
-0.03
|
0.00
|
-0.01
|
0.06
|
-0.06
|
0.01
|
-0.04
|
Bold indicates statistical significance, where p < 0.05
|
This study had several limitations. First, the dynamic CT image quality prevented the direct image segmentation and SCS definition. Our scanner was limited to a spatial resolution of 0.625 x 0.625 x 2.5 mm and a longitudinal coverage of 40 mm, not enough to fully capture the MC1 bone; however, newer scanners have dynamic protocols with larger longitudinal coverage and improved spatial resolution (e.g., 160 mm longitudinal coverage, 0.625 mm3 spatial resolution). We utilized HR-pQCT scans to obtain full joint images for initial image segmentation. However, this step poses challenges in vivo due to the long scanning time and increased exposure to ionizing radiation. Instead, a high-resolution static CT scan that includes the whole bone can be obtained from other clinical CT scanners that are widely available. For example, recent developments in cone beam CT scanners allows for imaging of the hand at resolutions up to 0.2 mm3 (e.g., CurveBeam HiRise weight-bearing CT scanners), which may be sufficient for providing a more detailed initial segmentation than images obtained from dynamic CT. Despite the additional image processing steps used here to overcome the scanner limitations, the TRE results indicate there was no substantial impact on joint angles. Second, the TMC joints in the acquired specimens varied in joint condition. Some specimens had considerable degeneration (e.g., joint space narrowing, presence of osteophytes, etc.) and others appeared healthy. We found that joint angle RMSE results were as high as 5.33° and 4.25° for inter-rater and intra-rater abduction-adduction, respectively (Fig. 8). However, the degree to which error is influenced by differences between healthy and diseased joints requires further study.