C

FIGURE 8 Some examples of CT-MRI registration. (A) MRI slice of a brain tumor patient with an inset of matched CT. The MRI clearly shows abnormal anatomy around the operation bed, while in the CT scan the brain around the operation bed (the dark spot) looks normal. In this case, the MRI is essential for accurate targeting of the tumor volume for radiotherapy. (B) Quality assurance of CT-MRI matching of the pelvis. The CT bone is overlaid in white over the MRI (bone is black in MRI). (C) Matching the same data without first erasing the femurs (Fig. 6a) leads to a grossly incorrect match (target registration error of about 1.5 cm).

Registration of CT with SPECT

The use of chamfer matching on SPECT images has been investigated for the brain to study Alzheimer's disease [12] and for the lung to quantify the relation between radiation dose and lung function changes [9,21]. Essential for successful matching is that the SPECT scan shows adequate anatomy of interest. For example, HMPAO SPECT scans show the whole brain, while ventilation or perfusion SPECT scans show the whole lungs. Segmenting these organs is, however, difficult because it is not easy to derive heuristic rules to find the correct threshold. With

FIGURE 9 Example of observer variation (top to bottom) and modality deviation (left to right) for the delineation of the prostate. The MRI slices in this figure have been matched and resampled to correspond with the CT slices on the left. The difference between the observers is a few millimeters, while the same observer delineates a target area in MRI that is almost a centimeter smaller. See also Plate 73.

FIGURE 9 Example of observer variation (top to bottom) and modality deviation (left to right) for the delineation of the prostate. The MRI slices in this figure have been matched and resampled to correspond with the CT slices on the left. The difference between the observers is a few millimeters, while the same observer delineates a target area in MRI that is almost a centimeter smaller. See also Plate 73.

chamfer matching it is, however, easy to include the segmentation threshold in the optimization procedure. This means that the whole chamfer matching algorithm is performed several times for different SPECT segmentation thresholds and that the match with the lowest cost function is kept. Because of the low resolution of the SPECT and PET, a single chamfer matching run is extremely quick and the whole procedure takes no longer than a minute. It is, however, essential in this procedure to use images with a calibrated scale and to exclude scaling in the optimization loop (changing the segmentation threshold for SPECT leads to a pseudoscaling due to the unsharpness of the image, which would be erroneously corrected by the matching algorithm). Mangin et al. describe a similar use of chamfer matching for MRI-PET image registration [23]

Measurement of Organ Motion

The use of chamfer matching to determine organ motion is illustrated in Fig. 10. Here, a pair of CT scans was first matched on the solid bone of the pelvis, and next on the right femur. The small insets on these figures show the drawing points used for matching. Erasing both femurs or erasing everything except a single femur takes about 15 seconds. To compute the relative motion (in this case of femur relative to pelvis), the transformation matrix of one match is inverted and multiplied with the matrix of the other match. The resulting matrix (in homogeneous coordinates) is decomposed into translation and rotation elements, which accurately quantify the motion of the femur relative to the pelvis. In this example, the femur was rotated by an extreme amount of 18.9° around the cranio-caudal axis and by smaller angles around the other two axes.

FIGURE 10 (A) CT-CT matched on the pelvis (the inset shows the contours used for matching, i.e., the pelvis only). (B) CT-CT matched on the right femur (the inset shows the contours of the single used femur).

FIGURE 10 (A) CT-CT matched on the pelvis (the inset shows the contours used for matching, i.e., the pelvis only). (B) CT-CT matched on the right femur (the inset shows the contours of the single used femur).

In a similar way, the motion of the prostate relative to the pelvis was quantified [35]. For matching the prostate, however, outlines delineated on the treatment planning system were used in the chamfer matching algorithm. Because all organ contours are used for matching, small differences in delineation of the prostate, missing slices, or differences in slice distance have only a limited influence on the accuracy. A strong correlation was found between rectal volume and anterior-posterior translation and rotation around the left-right axis of the prostate (Fig. 11). Because the rectum volume varies significantly, these parameters had the largest standard deviations of 2.7 mm and 4.0° (Table 2). Bladder filling had much less influence. Less significant correlations were found between various leg rotations and pelvic and prostate motion and bladder filling. Standard deviations of the rotation angles of the pelvic bone were less than 1 degree in all directions. The availability of statistical data on organ motion is extremely important in radiotherapy to decide on required safety margins.

Follow-up studies

By integrating functional lung images made before and after radiotherapy with the planning CT, the local relation between change of function and delivered dose can be quantified accurately [10]. The technique of chamfer matching is a convenient and more accurate alternative for the use of external markers on CT and SPECT [21]. In some cases, late damage visible in diagnostic scans can be related to local radiation dose, improving follow-up diagnostics. Figure 12a shows an example of tumor regression over a period of 3 months in a patient which was treated stereotactically twice at Harvard Medical School. Figure 12b shows CT scans of a patient with a lung reaction half a year after receiving a mantle field treatment. By correlating the planning CT with the follow-up CT based on the lung tops (the scans were made with the arms in different orientations), the given dose at the point of the lung reaction could be estimated. These data were important for the subsequent clinical decision, because it showed that the lung damage was most likely related to the high local radiation dose.

3 Performance Tests of the Chamfer Matching Algorithm

The performance of an automatic image registration algorithm may be quantified according to the following types of

FIGURE 11 Correlation between rectal volume differences (measured with the planning system) and rotation and translation of the prostate (measured using chamfer matching). Increasing rectal filling tends to rotate the prostate around a left-right axis of the patient near the apex of the prostate. The translations are measured because our reference point is the center of gravity of the prostate. Reprinted from Int JRadiat Oncol Biol Phys, 33, van Herk M, Bruce A, Kroes S, Shouman T, Touw A, Lebesque JV. Quantification of organ motion during conformal radiotherapy of the prostate by three-dimensional image registration, 1311-1320, copyright 1995, with permission from Elsevier Science.

FIGURE 11 Correlation between rectal volume differences (measured with the planning system) and rotation and translation of the prostate (measured using chamfer matching). Increasing rectal filling tends to rotate the prostate around a left-right axis of the patient near the apex of the prostate. The translations are measured because our reference point is the center of gravity of the prostate. Reprinted from Int JRadiat Oncol Biol Phys, 33, van Herk M, Bruce A, Kroes S, Shouman T, Touw A, Lebesque JV. Quantification of organ motion during conformal radiotherapy of the prostate by three-dimensional image registration, 1311-1320, copyright 1995, with permission from Elsevier Science.

TABLE 2 Magnitude of relative organ motion in the pelvic area specified as one standard deviation"

Relative motion Axis

0 0

Post a comment