Image Registration for Treatment Planning

A critical step in the radiotherapy procedure is the delineation of the target volume. Since the introduction of CT, the use of this image modality has become standard practice in precision radiotherapy. The CT scan has a high geometric accuracy, and the pixel values in the CT data are used directly to compute transport of the beam through the patient. Using a computer system, an optimal beam arrangement is searched to obtain a high dose in the tumor, while sparing critical normal tissues as much as possible (treatment planning). However, a disadvantage of CT is that the contrast between tumor tissue and normal structures is often extremely low. By using image registration, the diagnostic quality of, e.g., MRI can be combined with the geometric accuracy of CT [31].

Chamfer matching is highly suitable for 3D image registration because of its high speed. The extension of the chamfer matching algorithm to 3D is trivial. However, often the pixel size and the slice distance are different, which has to be taken into account in the algorithm. Using world coordinates in the matching algorithm as much as possible, limits the complexity of such an adaptation. In the next sections, the use of chamfer matching 3D registration of CT, MRI, and SPECT is described for the head, the pelvis, and the lungs.

CT-CT Registration

The registration of CT with CT is useful for treatment planning since it allows repeated measurement of the position of an organ, to quantify its mobility, or to update the knowledge of the anatomy of the patient during the course of treatment. In radiotherapy, the bony anatomy is generally used as a frame of reference, since the bony anatomy is most visible in the radiographs used for patient positioning. Segmentation of bone from CT is trivial, i.e., using binary segmentation at a suitable threshold level. The choice of threshold level is not very important, as long as the same threshold is used in both images to be matched. In general, however, a somewhat high threshold (e.g., 500 HU) is preferred to suppress the partial volume effect and to prevent erroneous identification of contrast dyes administered to the patient to enhance delineation of soft tissue structures. Applying a contour tracer to one

FIGURE 5 Use of chamfer matching for anatomy matching in electronic portal imaging. (A) Result of top-hat filter on the portal image shown in Fig. 3: bone is enhanced, but so is some noise. (B) Automatic segmentation of A. (C) Distance transform of B overlaid with a drawing of the anatomy in the reference image prior to matching. (D) After matching. Even though the quality of the segmented anatomical features is poor, the matching performance is reasonable.

FIGURE 5 Use of chamfer matching for anatomy matching in electronic portal imaging. (A) Result of top-hat filter on the portal image shown in Fig. 3: bone is enhanced, but so is some noise. (B) Automatic segmentation of A. (C) Distance transform of B overlaid with a drawing of the anatomy in the reference image prior to matching. (D) After matching. Even though the quality of the segmented anatomical features is poor, the matching performance is reasonable.

segmented image (e.g., slice by slice), and a distance transform to the other produces good quality input for the chamfer matching algorithm (Fig. 6A).

A problem that needs to be addressed, however, is that not all bony anatomy is rigid. In particular the lower jaw for the head and the leg bones for the pelvis may be in different positions in subsequent scans. To exclude these structures in our matching application, we have developed a small contour wiper that works on a wire-frame display of the contoured bones [37]. It turns out that erasure of the mobile anatomy is only required in the drawing input of the chamfer matching algorithm, since near the correct match the extraneous structures in the distance transform will not correspond with any structures in the drawing and will not be "seen" by the algorithm. With the mobile anatomy erased, CT-CT registration can easily reach subpixel accuracy [36].

A similar algorithm has been used to register CT-CT on lung contours to localize lung damage after radiotherapy [21].

CT-MRI Registration

Registration of CT and MRI is highly important in treatment planning of radiotherapy. CT is commonly used to define the organs and location of the tumor (target). CT is most suitable, since the pixel values in the CT scan provide a direct measure of the electron density of the tissue, information that is required to predict the transport of the radiation through the body. In addition, the good geometric fidelity of CT helps in delivering the treatment beams to the correct place. In some cases, however, MRI provides much better contrast of the tumor. In such cases CT-MRI registration is extremely useful. CT is then used to define the normal structures, while MRI is used to define the tumor. Bone is the most suitable structure for matching for treatment planning: first, because bone has a relatively high contrast in both CT and MRI, and second because bone is considered as frame of reference, relative to which all internal and external coordinates are related. In CT-MRI registration, it is best to use CT to extract the drawing, i.e., by binary segmentation followed by contour tracing. Also, for CT-MRI registration it is highly important to exclude mobile anatomy from the segmented CT, e.g., using an eraser tool such as shown in Fig. 6. Segmentation of bone in MRI requires selection of dark structures. For head matching, good results have been obtained with a two-threshold segmentation algorithm on the unprocessed pixel values [34]. For example, for proton density MRI, segmenting all pixels between 15% and 65% of the average gray value of the brain results in a reasonable segmentation of the skull, even though some parts of the brain and an outline of the skin are segmented as well (Fig. 7A). For the pelvis, application of a top-hat filter on the negated pixel values results in a reasonable segmentation [39] (Fig. 7B).

Some examples of CT-MRI registration are shown in Fig. 8. Note the strongly improved soft tissue visualization in MRI compared with CT. The chamfer matching algorithm is in widespread clinical use for CT-MRI registration for stereotactic radiotherapy [20]. The utility for conformal radiotherapy is great as well [29]. The importance of excluding mobile anatomy for matching is illustrated in Fig. 8C.

The impact of image registration on prostate treatment planning is dramatic. In a multiobserver, multimodality study, Rasch et al. compared the target delineation performed by three observers in CT, axial MRI, coronal MRI, and sagittal MRI [30]. If different observers delineate the target in the same scan, the differences are on the order of a few millimeters both for CT and MRI. However, between CT and MRI large systematic difference can be observed (Fig. 9). The delineated volume in MRI is on average 30% smaller. Most of these differences occur at the posterior side of the prostate (8 mm average difference). Even though the true target volume is unknown, it is likely that the MRI defined volume more closely represents the true organ, because in CT the base of the seminal vesicles was erroneously included.

Seminal Vesicle Removing

FIGURE 6 Use of a simple eraser tool to remove mobile anatomy for chamfer matching for CT-CT and CT-MRI registration. (A) Automatically traced bone contours of a pelvic CT scan. (B) The operator erases the femurs in the 2D picture. The operation is linked to the 3D point list, meaning that all slices are "cleaned" in a matter of seconds.

FIGURE 6 Use of a simple eraser tool to remove mobile anatomy for chamfer matching for CT-CT and CT-MRI registration. (A) Automatically traced bone contours of a pelvic CT scan. (B) The operator erases the femurs in the 2D picture. The operation is linked to the 3D point list, meaning that all slices are "cleaned" in a matter of seconds.

FIGURE 7 Segmentation of MRI for chamfer matching (single slice of the full stack is shown only). (A) Segmentation of the skull in MRI of the head using a two-threshold scheme. (B) Segmentation of bone in MRI of the pelvis using a top-hat filter on the negated image. The segmentation is good enough for successful matching.

Segmentation Breast Cancer
0 0

Post a comment