1887
banner image
No data available.
Please log in to see this content.
You have no subscription access to this content.
No metrics data to plot.
The attempt to load metrics for this article has failed.
The attempt to plot a graph for these metrics has failed.
A frequency-based approach to locate common structure for 2D-3D intensity-based registration of setup images in prostate radiotherapy
Rent:
Rent this article for
USD
10.1118/1.2745235
/content/aapm/journal/medphys/34/7/10.1118/1.2745235
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/34/7/10.1118/1.2745235

Figures

Image of FIG. 1.
FIG. 1.

Projection images of bone and non-bony anatomy. The images were computed from a kV CT of the pelvic region of a patient under an identity transformation. The CT was not part of the data set employed for the subsequent pose determination experiments. The CT was free of reconstruction artifacts and was considered to provide an accurate representation of the patient anatomy.

Image of FIG. 2.
FIG. 2.

Power spectral density and bone-to-bone spectral ratio (BNBSR). (a) The power in the bone and no-bone images illustrated in Fig. 1 as a function of spatial frequency. In both images, power decreases with frequency. (b) The BNBSR, expressed in dB, for different spatial frequencies. The BNBSR was computed as the ratio of the power spectral density of the bone image to the power spectral density of the no-bone region. We note the following three features of the BNBSR. First, the signal due to bone is considerably lower than no-bone at very low spatial frequencies. Second, there is a steep transition to much higher BNBSR at approximately . Third, BNBSR falls off to a plateau from the transition frequency to higher frequencies.

Image of FIG. 3.
FIG. 3.

Effect of high-pass filtering the DRR and portal image of the phantom. (a), (b), and (c) show a kV DRR at the identity pose, the DRR obtained after filtering with a high-pass FIR filter at a cutoff of , and the residual of the filtering operation, respectively. (d), (e), and (f) show the effect of the filtering operation on the corresponding portal image. There was no displacement between the two images. After filtering, bony anatomy in the DRR and portal image becomes clearer. (c), (f) Visual inspection of the residual DRR and portal image shows that the pixel intensity variations do not match exactly, and there is but a low degree of structural similarity between the two residual images.

Image of FIG. 4.
FIG. 4.

Effect of high-pass filtering on the objective functions. (a), (c) The 2D objective functions defined obtained by calculating correlation coefficient and mutual information between a kV DRR and the reference portal image over translations ranging from along the and axes. The correlation coefficient (a) did not peak at the correct pose of and , but was maximum at and . That is, the maximum was obtained when the DRR was shifted by along the positive axis with respect to the portal image. For the mutual information measure (c), although the global peak was at and , another maximum was observed at and . The objective functions defined by the correlation coefficient and mutual information were recomputed on the filtered DRR and portal image and are shown in (b) and (d), respectively. For both measures, filtering gave rise to well-behaved objective functions with the global peak at the correct position. This suggests that the bias and variance in registration accuracy can be reduced by improving bony anatomy contrast and reducing dissimilar intensity variations in the images being registered.

Image of FIG. 5.
FIG. 5.

Effect of high-pass filtering the patient images. The original kV DRR of a patient, computed at the identity pose, and an MV portal image of the same patient are shown in (a) and (d), respectively. The corresponding high-pass filtered images (cutoff of ) are given in (b) and (e) and the residuals of the filtering operation in (c) and (f). As with the phatom images, filtering improves the contrast of bony anatomy while removing dissimilar intensity variations in the DRR and portal image.

Image of FIG. 6.
FIG. 6.

Example of overlaid line-enhanced DRR and line-enhanced portal image prior to registration in the AP and LR view (a), (d) and after registration of the high-pass filtered AP images shown in Fig. 5 with the correlation coefficient (b), (e) and mutual information (c), (f). In these images, red represents linear features in the DRR and green represents linear features in the portal images. At alignment red and green features overlap to maximize the amount of yellow present in the fused images. It should be noted that although the registration was performed with the AP portal image only, both the LR and AP images were employed for visual validation of the results to ensure that the out-of-plane parameters had been correctly estimated.

Tables

Generic image for table
TABLE I.

1 through 15 denote the poses tested for with the pelvic phantom. The true transformation parameters , estimated transformation , magnitude of the displacement corresponding to , and overall error in estimating are shown for the correlation coefficient and mutual information. The overall registration was not correlated to the distance . Filtered images were employed in the registration.

Generic image for table
TABLE II.

Correlation coefficient and filtered images experiment results on phantom data. The table shows the mean, standard deviation, maximum and minimum value of the absolute value of the registration errors for each of the six transformation parameters and the overall registration error, , for poses 1 through 15.

Generic image for table
TABLE III.

Mutual information and filtered images experiment results on phantom data. The table shows the mean, standard deviation, maximum, and minimum values of the absolute value of the registration errors for each of the six transformation parameters and the overall registration error, , for poses 1 through 15.

Generic image for table
TABLE IV.

Estimated shifts in a patient’s position relative to the treatment field for correlation and mutual information-based registration over seven weeks. Only portal images in the AP view were used for the registration although images in both the AP and LR views were employed for visual verification of registration performance as illustrated in Fig. 6.

Loading

Article metrics loading...

/content/aapm/journal/medphys/34/7/10.1118/1.2745235
2007-06-26
2014-04-19
Loading

Full text loading...

This is a required field
Please enter a valid email address
752b84549af89a08dbdd7fdb8b9568b5 journal.articlezxybnytfddd
Scitation: A frequency-based approach to locate common structure for 2D-3D intensity-based registration of setup images in prostate radiotherapy
http://aip.metastore.ingenta.com/content/aapm/journal/medphys/34/7/10.1118/1.2745235
10.1118/1.2745235
SEARCH_EXPAND_ITEM