| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Original Research |
1 Department of Nuclear Medicine, Rouen University Hospital Charles-Nicolle and
Henri Becquerel Center, Laboratoire Universitaire QUANT.I.F., Rouen,
France.
2 Department of Radiation Therapy and Radiophysics, Henri Becquerel Center,
Laboratoire Universitaire QUANT.I.F., Rouen, France.
3 GE Healthcare, Israel.
4 Service of Radiology, Henri Becquerel Center, Laboratoire Universitaire
QUANT.I.F., Rouen, France.
Received October 25, 2004;
accepted after revision December 14, 2004.
Address correspondence to F. Gutman
(fabricegutman{at}yahoo.fr).
Abstract
 Top Abstract IntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1: Description of...References |
|---|
MATERIALS AND METHODS. The reproducibility of markers selection and the robustness of the method were assessed on a torso phantom. The accuracy of registration regarding the number and location of markers and the breathing state during CT was evaluated on eight patients using 10 external markers placed around the thorax before SPECT and CT acquisitions. The accuracy of registration was assessed using the mean errors (ME) between 10 markers after registration.
RESULTS. Registration using external markers on a phantom was accurate (ME, < 3 mm) when rotation was less than 40° (p = 0.02). The accuracy of registration improved markedly from four to six markers for phantom (5.5-3.6 mm) and patients (11.2-9.5 mm) and then remained constant up to 10 markers. The ME was less when using markers that well encompassed the thorax for phantom and patients (p = 0.02 and p = 0.05, respectively). The use of four anatomic markers was not accurate (ME, 20 mm).
CONCLUSION. The registration method is reproducible and accurate, and six external markers were required to get an ME of less than 10 mm in patients.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1: Description of...References |
|---|
|
|
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1: Description of...References |
|---|
Acquisitions—A dual-headed SPECT system (DST-XLi, GE Healthcare) equipped with two rotating parallel detectors and low-energy high-resolution collimators was used. The full width at half maximum (FWHM) of the system was 4.2 mm for a source-collimator distance of 5 cm. Sixty-four projections were acquired for 20 sec each and reconstructed using filtered back-projection with a Metz spatially varying filter (point spread function, 3; order, 4). The pixel size using a 128 x 128 matrix was 4.55 mm. Helical CT (Sytec+, GE Healthcare) of the phantom was performed on the same day. Sixty 512 x 512 slices resulting in a 0.9-mm pixel size were acquired. Helical CT images were obtained using a helical pitch of 6, 5-mm section thickness and intervals, 140 kV, and 220 mA.
Shift of phantom acquisition—The reconstructed SPECT phantom acquisition (reference volume) was used to generate 18 different phantom volumes (target volumes). The position of the phantom was not changed during the data acquisition, but shifts were applied by computer to the SPECT reference volume on a SPECT workstation (Vision, GE Healthcare). The reference volume was shifted successively on each of the three axes (x, y, z) using different rotations (10°, 20°, 30°, or 40°) or translations (5 or 15 pixels).
Image registration—CT and SPECT volumes were transferred for registration on a workstation (eNtegra ImageRegistration, GE Healthcare). CT data were reduced from a 512 x 512 format to a 128 x 128 format by pixel averaging. The registration method was landmark-based and consisted of identifying the same spot (external markers) on the reference and target images. The registration method is detailed in Appendix 1.
Factor of merit (FOM)—Spatial coordinates (cartesian reference coordinate system) of pairs of landmarks after registration were calculated from the spatial coordinates before registration and from the registration matrix. The registration matrix was obtained directly by the registration software after selection of pairs of markers. Specific software (ReadRGS, GE Healthcare) provided spatial coordinates of markers for each technique before registration.
To quantitatively evaluate the accuracy of registration for external
markers, we calculated the mean square errors, expressed in millimeters
squared (mm2), between the 10 homologous external markers after
each registration study. This criterion is called
"FOM10" and is calculated as follows:
 |
where Epp is the error per point.
Data analysis—First, two observers registered the 18 shifted volumes (target volumes) two times each with the reference SPECT using 10 external markers. The number of selected markers was 360 for each observer. Validation of the registration method was determined from the reproducibility of markers selection for intra- and interobserver variability and the robustness of the method based on the accuracy of registration variations when the target phantom was shifted.
Second, the optimal parameters for SPECT/CT registration were sought with various combinations of markers: number of markers from four to 10 and selection order of external landmarks using either selection order A, which used the four anterior markers first then lateral and finally posterior markers, or a second order B, which encircled well the thoracic volume with from four to 10 markers (Figs. 2A, 2B, 2C, and 2D).
Patient Studies
Patients—Registration investigations were performed on eight
patients with lung cancer (six men, two women; mean age, 66 years; range,
51-75 years). Table 1 shows the
patients' characteristics. All patients were recruited before beginning
regular radiation therapy treatment in the course of their disease. SPECT
scintigraphy with 99mTc-macroaggregate albumin and thoracic CT were
routinely performed to determine target volumes for radiation therapy.
|
External markers—Markers were fixed to the skin as follows: two on the sternum (markers 1 and 2), two on the nipples (markers 3 and 4), four lateral markers on the midaxillary line (markers 5-8), and two posterior markers on the inferior angle of the scapula (markers 9 and 10). Markers were placed before SPECT and were kept in place until the end of CT examination.
Acquisitions—SPECT and CT acquisitions were performed with the patient in a treatment position that was previously defined in radiation therapy. Patients were placed in a positioning system (Posirest, Sinmed BV) with arms extended behind the head. Great care was taken that patients were in the same position during both examinations. Sixty-four 20-sec projections were performed after injection of 150 MBq of 99mTc-macroaggregated albumin (Pulmocis, Schering) for SPECT data. Raw data were reconstructed by filtered back-projection using a Metz spatially varying filter (point spread function, 3; order, 4). The pixel size using a 128 x 128 matrix was 4.55 mm. Helical CT of the thoracic region was performed approximately 30 min after SPECT acquisition. Two sets of images were acquired after IV bolus injection of contrast medium. For each patient, 50-60 slices with a 512 x 512 matrix resulting in a pixel size of 0.9 mm and slice thickness of 5 mm were acquired. Two sets of images were consecutively acquired. The first set (CT1) was acquired during normal breathing and the second set (CT2), during breath-holding after deep inspiration. Data sets of CT1/SPECT and CT2/SPECT were registered using either external or internal markers.
FOM and data analysis—FOM10 was used to quantify registration quality using external markers. One of the eight patients (patient 7 in Table 1) was excluded from this study because of a major location variation for one external marker between SPECT and CT acquisitions. The optimal parameters for SPECT/CT registration were sought with various combinations of markers: number of external markers from four to 10 and selection order of external landmarks using either selection order A or B (Figs. 2A, 2B, 2C, and 2D), breathing state during CT for patient's data (deep inspiration or free breathing), and registration using the four internal landmarks. We visually selected four internal markers that could be seen on SPECT and CT data: two pulmonary apices and two costophrenic angles. The registration accuracy with internal landmarks was then evaluated using the mean error (ME) of six external markers (markers 1, 3, 6, 7, 8, 9) (FOM6) and was compared with the results of registration using four external markers (markers 2, 4, 5, 10).
|
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1: Description of...References |
|---|
Robustness of the registration method— The robustness of the registration method was evaluated on the basis of the behavior of the method when a shift was applied to the target volume. Figure 4 shows the results of FOM10 according to the transformation. Quality of registration was not influenced by the transformation: it was either a rotation (p = 0.12) or a translation (p = 0.99). FOM10 was similar with a rotation angle of 10-30°. Only a 40° rotation provided a significantly less accurate registration than 10° (9.7 ± 1.3 vs 6.7 ± 2.1 mm2, respectively; p = 0.02).
|
|
The effect of selection order was also significant. The mean FOM10 was 15.3 ± 12.1 mm2 for A and 10.7 ± 3.8 mm2 for B (corresponding to mean point errors of 3.9 and 3.3 mm, respectively). When selecting markers with selection order A, FOM10 decreased markedly from four to six markers and then reached almost a constant value up to 10 markers. When selecting markers with selection order B, FOM10 decreased between four and five markers (from 19.0 to 10.5 mm2) and then decreased much slower.
Patient Results
Perfusion lung SPECT—Five patients did not undergo surgery
before radiation therapy. SPECT lung scans showed marked heterogeneity in the
peritumoral area in four patients (patients 4, 5, 6, and 8) compared with the
contralateral perfusion. For patient 3, lung perfusion was normal around the
tumor. Figures 6A, and
6B shows two cases of lung
perfusion around the tumor (patients 3 and 4).
External markers—Figure 7 shows quality of registration (FOM10) plotted versus the number of markers and the selection order for seven patients under free breathing during CT. The mean FOM10 was 88.4 ± 7.7 mm2, corresponding to an error per point of 9.4 mm. FOM10 decreased markedly between four and six markers (from 125.9 to 91 mm2) and then decreased much slower until 10 markers (64 mm2).
|
Internal landmarks—Reproducibility for selection of internal landmarks was evaluated both on CT and SPECT. One patient was excluded from the study because of a pneumectomy. The MEs for intraobserver variability were 1.03 ± 0.11 pixels and 1.47 ± 0.26 pixels for the two observers. The ME for the interobserver variability was 1.80 ± 0.20 pixels. Quality control of registration was evaluated by computing FOM6. The mean FOM6 was significantly higher using four internal landmarks (320 ± 47 mm2) than using four external landmarks (162 ± 50 mm2) (Fig. 8), corresponding to mean point errors of 18 and 12.7 mm, respectively (p = 0.007).
|
|
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1: Description of...References |
|---|
These results are in agreement with those of previous studies that showed registration quality improving with the number of markers [11] and markers that encompass well the volume of interest [12]. The 3-mm registration error was considered accurate for both diagnostic and therapeutic applications; thus, this software could be used for rigid studies such as brain registration.
The main results of the patient studies were, first, that six or more external markers with two lateral or posterior markers were needed to have a registration accuracy of less than 9 mm. Second, selection series B, which well encompassed the thoracic volume, provided more accurate values than series A, but primarily when the number of markers was fewer than six. This was not surprising because selection orders A and B converged with an increasing number of markers, and using six markers with A appropriately covered the thoracic volume. Third, the use of internal markers did not provide accurate registration and had poor to bad reproducibility. The choice of internal landmarks was hindered by the limited number of anatomic data visible on lung SPECT. Selection of internal landmarks on the two apices and costophrenic angles was not suitable for patients who underwent lung surgery or who had a large tumor at the apices or bases.
The difference in the registration quality of external markers and internal landmarks in our study was in agreement with Forster et al. [13] who showed in 10 patients significant differences between external and internal markers (ME, 2.36 ± 1.89 mm for internal markers and 1.72 ± 0.95 mm for external markers). The number of internal markers needed to obtain a good registration was probably higher than the number of external markers. Maguire et al. [14] used internal landmarks to register fluorine-18-fluorodeoxyglucose (18FDG) PET and thoracic CT and showed that a number of seven to 15 internal markers would be more appropriate. However, some authors obtained good registration using internal landmarks. Perault et al. [15] reported registration between SPECT and thoracic CT using internal landmarks. When the sternum and clavicles were used as landmarks for thorax, the quality of registration was inferior to the pixel size using a 64 x 64 matrix (7 mm). Birnbaum et al. [16] also performed abdominal registration using internal landmarks and obtained a registration error of about 1.5 pixels.
Limitations of the Study
External markers—Special attention was focused on the design
and fixing of external markers. The shape of the external markers was designed
so that the radiolabeled solution was placed as close as possible to the lead
sphere, and the markers were fixed to the skin in order that they would not
need to be replaced between SPECT and CT acquisitions. However, markers appear
larger on SPECT images than on CT images, and a lower activity of
99mTc would have been more appropriate, such as 0.1 MBq in 75 µL
per marker, as described by Marks et al.
[17]. One of the drawbacks of
registration using external markers was caused by the movements of the markers
between the two acquisitions, primarily due to arm position within the
constraining system. In our experience, these skin movements occurred mostly
from markers fixed to the scapula. Unless a good constraining system was used,
posterior external markers should have been placed in an area of less
movement.
Quality registration control—Even if validation of registration methods was previously tested by visual analysis with success [18, 19] on CT and MRI studies, the use of an objective criterion was required when low-resolution data were used (e.g., SPECT studies). The most popular registration quality criterion for rigid methods was the Euclidian distance between reference and target markers after registration [20, 21]. Several indexes were used such as the fiducial registration error [20, 21], which corresponds to the mean variation of coordinates between homologous markers after registration. The quality criterion we used was a similar criterion called "FOM10," which corresponded to the mean square errors between all external markers after registration, whatever the number of markers used for registration. This criterion was not optimal because it was partly dependent on markers used for registration. It would have been better to use different markers for registration and for quality control to avoid mistakes due to marker selection. Because the maximum number of markers for our registration software was 10, we could not select different markers for registration and for control. This was possible only for quality control of internal markers (FOM6) using six external markers for quality control that were not used for registration with either internal landmarks or external markers.
Breathing artifacts—A registration accuracy of 9 mm on patient data was probably enough for diagnosis if CT was used only for anatomic location; however, for radiation therapy and therapeutic applications, the required accuracy of registration should have been higher. The main source of error in our study was that the thorax was not a motionless organ because of breathing. The MEs were higher when CT was acquired during deep inspiration, especially when using internal markers. Yet, there was no significant difference in our study between the two CT breathing states, probably because of lack of statistical power due to the small number of patients studied. Beyer et al. [22] showed that artifacts due to breathing for thoracic registration happened in 42 of 43 cases and that the best breathing state for CT acquisition would be breath-holding after normal expiration. Goerres et al. [10] showed that the registration errors for thoracic acquisitions varied from 6 to 20 mm with regard to the breathing state and were minimum with free breathing or breath-holding after normal expiration.
Even if in many institutions patients breathe freely during treatment planning studies under the assumption that this will average out the effects of breathing, the motion of the tumor and of the lung itself during the delivery of each treatment appears to affect the outcome of radiation therapy in patients with inoperable non-small cell lung carcinoma. Lung tumors have been shown to move substantially during quiet breathing, causing inaccuracies in treatment delivery [23, 24]. To compensate for this motion, we usually use a large margin, consequently increasing the amount of normal lung tissue in the high-dose volume.
Two distinct techniques have been used to reduce the effect of respiratory motion. The first involves confining the radiation delivery to a specified phase in the breathing cycle by gating the linear accelerator while the patient breathes freely. Breathing is monitored with devices that trigger radiation delivery during specific phases of the patient's respiratory cycle [25]. In a recent study, Suga et al. [26] showed marked improvement in registration quality on rigid thoracic registration using respiratory gating. Gated images yielded a significantly better SPECT-CT match compared with ungated images in seven patients (4.9 ± 3.1 mm vs 19.0 ± 9.1 mm), but the evaluation was visual and only one or two internal landmarks were used. Wong et al. [27] proposed this technique to decrease the safe lung volume from external beam radiation with promising results, and Stromberg et al. [28] showed a 12% decrease of lung volume irradiation with gating.
For the second approach, breathing is controlled either voluntarily by the patient, such as the breath-hold technique, or by using an occlusion valve. The use of respiratory-gating devices may be able to overcome breathing artifacts. Even if some authors showed advantages of a breath-hold technique for both CT and radiation therapy [29], SPECT/CT or PET/CT simulation both necessitate a free-breathing-gated technique because of the duration of scanning.
Use of a Rigid Registration Method
The use of a rigid registration method is more suitable for brain
registration and was theoretically not optimal for this application.
Nevertheless, rigid registration methods can probably be improved and still
remain a valuable tool even for thoracic registration. Furthermore, rigid
registration methods are today the only commercially available methods for
which the validation methods are well known. Validation of rigid methods
either can use visual comparison of anatomic landmarks or can involve the use
of external markers [21,
30]. Nonrigid methods are not
yet commercially available, and no standard validation method has been
described [31]. The solution
of these contradictions may be the advent of dual-technique instrumentation
that has provided a direct fusion capability. In-line SPECT/CT systems have
been used both for diagnosis
[32] and for therapeutic
applications [33], especially
when SPECT provides few anatomic data, such as sestamibi, pentetreotide,
iodine-131, and, of course, lung perfusion imaging.
In conclusion, we performed phantom and patient studies to evaluate the use of rigid registration of lung SPECT and thoracic CT. The selection of markers was well reproducible, and the robustness of the software was good for rotations less than 40°. Our results showed that external markers provide better registration than internal markers. Five markers, including a posterior marker, were required to get a registration error of less than 3 mm on phantom study. Six markers, including two lateral or posterior markers, were required on patient data to get a registration error of less than 9 mm. These results confirm that the use of a rigid registration method—even when optimizing the number and location of external markers—is suitable for motionless organs but is not optimal for thoracic registration.
APPENDIX 1: Description of the Registration Method
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1: Description of...References |
|---|
In our study, data sets of SPECT/CT were registered using external markers.
The landmark-based registration method consists of identifying the same
anatomic spot on the reference and target images (a landmark is a pair of
graphical markers, one marker being placed on the reference image, while the
other is placed on the target image). At least three, but fewer than 10,
landmarks are required to compute a true 3D transformation. The computation of
the rigid-body transformation T is the result of applying an
optimization procedure minimizing the sum FOM(T,n) as
defined below:
 |
where n is the number of selected landmarks, T is the transformation we are looking for, and ERR(T,i) is the distance—reference marker landmark i, T(target marker landmark i) (hence, ERR(T,i) is the distance measured in the reference space between the reference marker and the transformation of the corresponding target marker). The optimization procedure is based on a genetic algorithm. To use a genetic algorithm, you must represent a solution to your problem as a genome (or chromosome). The genetic algorithm then creates a population of solutions and applies genetic operators such as mutation and crossover to evolve the solutions to find the best one(s).
Acknowledgments
We thank Richard Medeiros, Rouen University Hospital Medical Editor, for
his valuable advice in editing the manuscript.
|
TopAbstractIntroductionMaterials and MethodsResultsDiscussionAPPENDIX 1: Description of...References |
|---|
CiteULike 
Complore 
Connotea 
Del.icio.us 
Digg 
Reddit 
Technorati What's this?
This article has been cited by other articles:
|
|
 |
|
  K. Suga, Y. Kawakami, H. Iwanaga, O. Tokuda, and N. Matsunaga Automated Breath-Hold Perfusion SPECT/CT Fusion Images of the Lungs Am. J. Roentgenol., August 1, 2007; 189(2): 455 - 463. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 B. Harris, D. Bailey, S. Miles, E. Bailey, K. Rogers, P. Roach, P. Thomas, M. Hensley, and G. G. King Objective Analysis of Tomographic Ventilation-Perfusion Scintigraphy in Pulmonary Embolism Am. J. Respir. Crit. Care Med., June 1, 2007; 175(11): 1173 - 1180. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 J. Petersson, A. Sanchez-Crespo, S. A. Larsson, and M. Mure Physiological imaging of the lung: single-photon-emission computed tomography (SPECT) J Appl Physiol, January 1, 2007; 102(1): 468 - 476. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK |
