Table 1.
Patient characteristics.
Figure 1.
The procedure for the landmark-based registration accuracy measurements by DiffeoMap.
A: The 1st time point image with landmarks placed on the distal and proximal bifurcations of vessels (red arrows). B: The landmark coordinates placed on the first time point image were transferred to the linearly registered second time point image, revealing the degree of misalignment. C: Landmarks were manually relocated to the corresponding bifurcations in the 2nd time point images, and the distances of misalignment were measured.
Table 2.
Registration accuracy.
Figure 2.
Comparison of three registration results.
A: The original first time point image. B: Linearly registered second time point image. Note that only the region of interest was normalized in the second time point. C: A subtracted image between the first time point and the linearly registered second time point. D: A subtracted image with the first iteration of LDDMM (single α LDDMM) (α/γ = 0.01). E: A subtracted image with the cascading α LDDMM (α/γ = 0.01−0.005−0.002). F: A Jacobian map calculated from the transformation matrix of the cascading α LDDMM. Red and blue arrows indicate locations of growing tumors (red) and misalignment artifacts (blue). The dotted lines show reciprocal positions.
Figure 3.
Comparison of growing and emerging tumor.
A: The original first time point image. B: The linearly registered second time point image. C: A subtracted image between the first time point and the linearly registered second time point. D: A subtracted image with the first iteration of single α LDDMM (α/γ = 0.01). E: A subtracted image with the cascading α LDDMM (α/γ = 0.01−0.005−0.002). F: A Jacobian map calculated from the transformation matrix of the cascading α LDDMM. Orange, red, and blue arrows indicate locations of emerging tumor (orange), growing tumors (red), and misalignment artifacts (blue). Note that almost the entire misalignment is removed by LDDMM, clearly indicating the small nodule that appeared in the second image (orange arrow). This new nodule was not detected by the Jacobian map, which is a metric of growth.
Figure 4.
Temporal change in nodules and results of transformation in serial CT scans, with visual presentation in a subtraction image and a Jacobian map.
The subtraction image shows gray for pixel value = 0, black for negative, and white for positive value. The color-coded Jacobian map shows green for Jacobian <1, red for Jacobian >1, and yellow for Jacobian = 1.
Figure 5.
Automated and quantitative growth measurement using the cascading α LDDMM transformation matrix.
Linear registration (A) shows some mis-registration as black or white linear structures, whereas LDDMM transformation (B) shows complete nodule registration, as well as the surrounding lung parenchyma. Once the nodule at the first time point is defined (C), the nodule definition can be automatically transformed to the second time point (E), which enables automated volume measurement.
Figure 6.
Comparison between cascadingα LDDMM and B-spline registration.
A metastatic nodule in the right upper lung (red arrow) was growing slightly at clinical follow-up (A: 1st time point, B: 2nd time point). C: Subtracted image with B-spline registration shows slight mis-registration along the lung parenchyma and a thin, rim-like difference around the nodule. D: Jacobian map obtained from B-spline registration is inhomogeneous and volume expansion is not clear. E: Cascading LDDMM shows complete registration. F: Jacobian map from cascading LDDMM obviously shows a red-colored spot, which corresponds to the growing nodule.