Mammotomography with Pinhole Incomplete Circular Orbit SPECT

MATERIALS AND METHODS

A 3-head clinical gamma camera (Triad XLT-9; Trionix Corporation, Twinsburg, OH) with a 40-cm transverse × 20-cm axial field of view (FOV) and 9-mm-thick NaI(Tl) scintillator, was used for all experiments. Image data were thus acquired in 128 × 64 pixel matrices. The system allows for sufficient radial head retraction for patient PICO-SPECT imaging as well as posterior patient detector clearance. A 15%-wide symmetric energy window was used about the 140-keV photopeak from ^99mTc in all studies. The camera used a custom-designed and -built pyramidal collimator support with 14-cm focal length (27,28), which accepts a single and exchangeable pinhole insert shrouded by tungsten shielding. Originally intended to reduce penetration from higher energy γ- rays, the patient-side tungsten shielding secures the exchangeable pinhole inserts yet physically increases the minimum ROR by 4 cm, because it protrudes from the pinhole insert itself (28). Although it may otherwise present a problem for imaging very small volumes because of resolution degradation with the increased distance from the pinhole, this 4-cm distance actually facilitates keeping the relatively large breast phantom entirely within the approximately 90° aperture acceptance angle, and hence within the camera FOV, and also affects the maximum attainable orbital range.

RESULTS

The GeLi spectrum of the DU collimator measured up to 2 MeV yielded several peaks (Fig. 4A). These low-energy peaks in the GeLi spectra were also evident in the energy spectra measured with the gamma camera (Fig. 4B). The low-energy background peaks occur below but near the 140-keV photopeak in the ^99mTc scatter spectrum and increase in magnitude with smaller pinhole aperture. The peaks may be partly the result of characteristic x-ray fluorescence, which occurs at 115 keV in DU. By subtracting the tungsten energy spectra from the corresponding-sized DU spectra and integrating over a ±7.5% symmetric energy window about the photopeak used to collect the image data, the DU background contribution in the photopeak events was estimated (Table 2). The DU background in the projection images without additional emission events was spatially uniform, yielding decreased image contrast and had an average counting rate for all DU apertures of 62 counts/s over the entire 40 × 20 cm² camera FOV.

As expected, the reconstructed PICO-SPECT image data demonstrated that lesions with lower activity concentrations are more difficult to distinguish than those similarly sized and located lesions with higher activity concentrations (Figs. 2B and 5). For the larger lesion size, measured contrasts formed an extended peak around the 2-mm-diameter aperture, whereas contrast for the smaller lesion increased monotonically with decreasing aperture size (Figs. 5A and 5C). The measured contrast ranges varied from the known 7:1 lesion-to-background ratio (contrast = 6) by factors of 2.8 for the large lesion using a 1-mm-diameter aperture to 11.2 using a 4-mm aperture for the smaller lesion. In general, greater differences were apparent with the larger pinholes and smaller lesions. Similar results were obtained for lesion contrast with the 12:1 activity concentration ratios (contrast = 11), with the measured range of differences from the known activity concentration varying from factors of 2.4 for the large lesion using a 2-mm- diameter aperture to 8.0 using a 4-mm aperture for the smaller lesion. The mean contrast values between tungsten and DU collimator materials were nearly equivalent for the 7:1 and 12:1 concentration measurements. However, there were measured differences ranging from a factor of 1.4 for the 1-mm aperture and 12:1 lesion-to-background ratio to a factor of 2.5 for the 4-mm aperture and 7:1 ratio between contrast values with different lesion sizes. With increasing aperture diameter, improvements in SNRs in factors of 1.6–4.0 from the lowest to highest measured SNR in the set were observed for various lesion sizes and both activity concentrations and pinhole materials (Figs. 5B and 5D). The measured SNRs formed a broad peak between the 2- and 4-mm aperture diameters.

Results for the pinhole planar and reconstructed PICO-SPECT comparison demonstrate that PICO-SPECT images yield improved lesion contrast (Fig. 6). The peak-to-background heights (contrast) in the image profiles of the planar and PICO-SPECT images illustrate clear differences between acquisition approaches. Although data acquired for 1 aperture size with planar and PICO-SPECT imaging were collected for the same times, the collection time between aperture sizes was adjusted to collect approximately 560,000 total counts in both acquisition techniques. These count densities correspond to clinical count densities based on known lesion concentrations (1). In the planar image (Fig. 6A), the activity-filled myocardial insert as well as the acrylic boundary between the breast and torso phantom are clearly visualized; however, visualization of the smaller 0.6-cm lesion is difficult. The myocardial apex is barely seen at the image left in the reconstructed PICO-SPECT sagittal image (Fig. 6A) because of the zoom used in the reconstruction and because myocardial sampling and distance varied with projection angle, affecting the counts seen from the myocardium as the camera traversed the incomplete circular orbit about the breast.

The reconstruction of the PICO-SPECT acquisition about the cold minidisk phantom inside the radioactively filled breast phantom illustrates both sampling uniformity and depth limitations of this breast imaging approach with a pinhole collimator (Fig. 3B). Under these measurement conditions, sampling is apparently approximately complete to about 7.5 cm from the nipple in the 12-cm-deep breast, as evidenced by the relatively distortion-free recovery of the disks to that depth. The sampling appears to be markedly incomplete past that depth from the nipple, but this still may not preclude adequate hot lesion imaging on nearly uniform backgrounds as evidenced in other approaches (15,16,20).

The images obtained from the initial patient study (Figs. 7 and 8) demonstrate the improvements observed between planar and dedicated PICO-SPECT phantom studies. PICO-SPECT acquisition times were, however, 4 times longer than the planar acquisition. A lesion was clearly seen with both acquisition methods, but the coronal and sagittal slices through the reconstructed PICO-SPECT data revealed decreased intensity in the center of the tumor. The small (approximately 3-mm) distributed point sources can also be seen sporadically (although with some streak artifacts) in the PICO-SPECT images but are not visualized at all in the planar pinhole image. Mean contrast measurements from the sagittal and coronal views, with annular ROIs to account for the center of the lesion with decreased intensity, were measured to be 6.4 times higher with PICO-SPECT than with planar pinhole imaging. The absolute contrast was 11.6, which is 2.5 times greater than the best contrast measured with the phantom-simulated 12:1:20 uptake ratio in the larger tumor with the 3-mm tungsten pinhole.

DISCUSSION

Several pinhole apertures from 1 to 4 mm and made with 2 different materials were investigated for use with dedicated PICO-SPECT imaging of the breast with ^99mTc compounds. DU was initially investigated as a candidate for higher resolution imaging because of its lower expected edge penetration compared with other collimator materials at 140 keV, but the results indicated that at this energy and these RORs, both DU and tungsten have similar performance in terms of SNR, contrast, and image quality. Thus, further comparisons of the tungsten apertures were made between the PICO-SPECT approach and planar pinhole imaging, and an optimal tungsten aperture was chosen for an initial clinical patient study.

Because of the nonstandard pinhole aperture material using DU, the intrinsic spectral emission of DU was measured. The presence of the peaks in the DU background spectra (Fig. 4) is caused by trace quantities of fissionable ²³⁵U and associated decay products. At medium to high energies (364 to 511 keV), for which these pinholes were originally designed, the continuum of events is most probably caused by β-decay-created bremsstrahlung, with no distinct or predominant transition energies. However, as observed at lower energies (Fig. 4A), there are some peaks, 1 of which may be the result of characteristic x-ray fluorescence that may affect imaging near 140 keV. Characteristic x-ray peaks clearly appear in the gamma camera-measured spectra with DU but not tungsten and, in fact, would otherwise be excluded with photopeak energy windowing.

The intrinsic background from DU was measured to be <12% of the windowed photopeak signal with the smallest aperture and decreased to approximately 5% contribution with the larger 2- to 3-mm apertures. This increased background observed with smaller apertures (Fig. 4B) is related to the relative decrease in pinhole sensitivity for emission events (Table 2) but correspondingly near-constant mass of DU material, hence constant background event rate. Also evident is the high-energy plateau beyond the ^99mTc emission photopeak due to the DU background, and counts in this plateau may partially account for the slightly higher measured sensitivities shown in Table 2.

No spectral degradation was observable with the lower-Z and -density tungsten apertures. Using the tungsten and DU pinhole linear attenuation coefficients to calculate the effective pinhole diameter with Anger’s (31) original formula at 140 keV, there should be a <0.1% difference between these pinhole materials for all aperture sizes and specific acceptance angles (Table 2). Thus, tungsten would be a sufficient collimator for high-resolution pinhole imaging at 140 keV without any additional background from the collimator itself.

Slightly higher than expected clinical count scans obtained with each pinhole were also investigated with lesions of different uptake ratios. The coronal views of these reconstructed data (Fig. 2B) demonstrate some truncation of the projection data in the superior-to-inferior (seen from right-to-left) aspects of the reconstructed breast data. This truncation is a potential problem with high-resolution pinhole imaging of the breast if magnification, which may truncate the projections, is used in the reconstruction. More importantly, if the patient’s breasts are large and positioning is difficult, truncation is likely. Planar breast imaging (pinhole scintimammography) alone can also be limited by partial truncation of the entire breast volume (22). Hence, care must be taken to ensure that the entire breast volume is encompassed throughout the trajectory and that any suspected lesion site is included. With pinhole imaging, there is distortion at the axial extremes of the image because of well-known incomplete sampling there. Ideally, some axial translation could be used to encompass large, pendulous breasts entirely in the FOV. Three-dimensional orbits that use continuous angular rotation with continuous axial bed motion may also be used to more completely sample the entire pendulous breast (18).

Smaller differences were noted between pinholes of different materials than between different aperture sizes (Fig. 5). The trends in lesion SNRs indicate that there are optimal pinhole apertures for different sizes of lesions in this dedicated breast imaging application, and these 2- to 4-mm values correspond with those measured in another tomographic technique (23). In addition, the application of attenuation and scatter compensation may further improve these relative results.

There is a well-defined separation in contrast values of large and small lesions (Figs. 5A and 5C) for various lesion uptakes. The overall results from these measurements are consistent with improvement in contrast as pinhole diameter decreases. For the 12:1:20 ratio and large lesion, the quadratic fit indicated a slight decrease in contrast from 2- to 1-mm-diameter aperture. This could be attributed to measurement noise. Furthermore, the contrast values for both DU and tungsten apertures are very similar for each aperture size, with the results using tungsten pinholes usually slightly better (+9%) than those with DU. This difference indicates that the background contribution from DU (62 counts/s) seems to play an insignificant role in signal degradation compared with the results with nonradioactive tungsten inserts in this application. Because DU is somewhat more difficult to obtain and, more important, to machine to a desired configuration, tungsten apertures would be better for this application.

The individual SNR results for tungsten pinhole collimators with various lesion uptake ratios are on average +20% better than those with DU apertures (Figs. 5B and 5D), indicating that the background from DU may degrade the lesion signal intensity for the same projection image noise. The differences between the mean SNRs measured from smaller and larger lesions with both material types and for both activity concentrations are considerably greater than the differences between SNRs for materials at a fixed aperture diameter. The quadratic parameter estimates for the large lesion yield a near-optimum aperture at 3.8 mm, whereas for the small lesions the near-optimum value is 2.6 mm. Thus, although pinholes with larger apertures would provide better sensitivity, their poorer spatial resolution ability apparently reduces SNR of smaller lesions or lesions with lower radioactive uptake in favor of larger lesions or lesions with greater uptake. These SNR results suggest that in an effort to detect earlier stage breast cancers, a small aperture (2–3 mm) demonstrates a good trade-off between resolution and sensitivity.

For comparison measurements between planar pinhole imaging and PICO-SPECT, smaller apertures outperformed larger apertures in absolute lesion contrast (Fig. 6B), generally agreeing with the earlier results (Figs. 5A and 5C). (For these measurements, SNR was not calculated; the measurements had nearly identical noise values for a given aperture diameter because the measurements were made for equal times. In this case, the SNR trends track contrast measurements.) Thus, the loss in sensitivity for SPECT with smaller, finer resolution, nonzero-diameter pinholes may be outweighed by the gain in spatial resolution.

Although the planar pinhole studies were not necessarily optimized, a range of aperture sizes was tested. Moreover, these results are corroborated by other recent breast measurements that indicate that pinhole tomographic imaging is superior to pinhole planar breast imaging (23).

Comparisons of the uncompressed breast with PICO-SPECT and planar data indicate that the tomographic acquisitions yield more than a factor of 20 improvements in contrast (Fig. 6B). Comparisons were made only for measured values from the large 1.0-cm lesion, because the small lesion was not observed in all the planar data, especially with the larger, otherwise more sensitive pinholes. For example, calculated contrast values obtained from these small lesion image datasets were large and negative, most likely because of differences measured between 2 small numbers.

Sampling limitations throughout the breast volume were illustrated. The posterior breast side of the most distal disk is 8.5 cm from the nipple. Although blurring is clearly evident at this depth (Fig. 3B), the cold disk up to 7.5 cm can be visualized. Full 360° circular orbit acquisitions perpendicular to the nipple-chest axis (i.e., standard horizontal AOR SPECT) through the breast have demonstrated adequate sampling with artifact reduction (10), but this implementation is not possible with patients or anthropomorphic torso phantoms without torso interference that necessitates very large RORs. Nevertheless, imaging of hot lesions with PICO-SPECT is still possible for expected focal radiotracer uptake and an otherwise uniform background. Insufficient sampling of this general paradigm has not been shown to be problematic in lesion visualization (15,16,20,21). Complete imaging of the axilla, on the other hand, may be considerably more difficult without the use of alternative or additional orbits (18,33,34). One group has successfully used a limited angular acquisition pinhole SPECT centered on the axilla and already obtained improved clinical sensitivity and specificity for lesion identification (25).

The primary reason PICO-SPECT and other SPECT imaging approaches may indeed provide improved SNR and lesion contrast is because of the application-specific nature of the imaging task for dedicated emission mammotomography. The current results imply that smaller breast lesions or lesions with lower uptake could possibly be detected with small-aperture, high-stopping-power collimators and PICO-SPECT acquisitions.

Images obtained from the initial patient scan illustrated the improved image quality and information content available with dedicated volumetric imaging with PICO-SPECT (Fig. 8). Although the angular acquisition range for the patient was smaller (by approximately 30°) than for the phantom scans because of physical limitations, the reconstructed images yielded additional information about the lesion not available with planar imaging. In this case, increased counts were clearly visualized in the lesion periphery. This patient received preoperative neoadjuvant chemotherapy consisting of doxorubicin and taxol, administered concurrently with hyperthermia. She had a complete pathologic response at the time of surgery.

This initial image set yields information about the viability of this imaging approach by demonstrating a 6-fold contrast improvement over the planar pinhole technique. Resolution limitations resulting from the size of the lesion played a smaller role in this patient scan than in the phantom measurements using smaller solid lesions, and the large lesion size may explain the very large absolute contrast. The tumor widths measured at the profile locations indicated in Figure 8 were all larger (4%–29%) than the largest (3.7-cm) lesion dimension as measured by ultrasound. Because single-photon image resolution degrades with increasing distance, the larger measured size with PICO-SPECT may result from the system resolution fall-off. Overall, 3-dimensional visualization and localization of the lesion, however, were considerably more clear with PICO-SPECT than planar imaging, in addition to the improved contrast.

With pinhole iterative reconstruction techniques modeling spatial resolution, the lesion size and contrast values could be expected to improve. Indeed, iterative reconstruction techniques would minimize image artifacts (9,18,24), and modeling the system response would improve image quality and lead to more accurate quantification overall.

There are, however, some caveats in PICO-SPECT imaging of the breast. Large breasts may not entirely fit into the FOV, and to adequately sample the breast volume without truncation, multiple axial scans may be needed. Three-dimensional orbits designed to satisfy complete sampling conditions may also be important, regardless of breast size (18,33,34). Large RORs may also be required for larger breasts, along with concomitant resolution loss with the increased distance. Scanning the disk phantom within the breast phantom volume demonstrated that there may be sampling insufficiencies at depth for large pendulous breasts. Additional scans with variations in tumor locations throughout the breast still need to be performed. Furthermore, imaging the axilla and tissue near the chest wall could be accomplished with additional static views of those regions.

CONCLUSION

We have demonstrated a new dedicated emission mammotomography technique with PICO-SPECT imaging of breast lesions of various sizes and activity concentration ratios, yielding images of small lesions (down to 0.6 cm) with high contrast and SNR. The technique is readily implemented on contemporary clinical SPECT systems equipped with pinhole collimators and appears to be a better technique than conventional planar pinhole scintimammography. Medially located phantom lesions were more easily seen with PICO-SPECT than pinhole scintimammography. However, evaluation with lesions located at several locations throughout the breast, near the anterior chest wall, and axilla are further warranted.

Both DU and tungsten pinhole collimators had nearly equivalent performance for the imaging task at 140-keV imaging of aqueous ^99mTc. Because of easier accessibility to tungsten materials and their fabrication, as well as the lack of additional radioactivity concerns, this high-stopping-power material at 140 keV is recommended along with this imaging technique. Furthermore, it was determined that 2- to 3-mm aperture pinholes provided the best SNR along with fine spatial resolution for smaller lesions and lesions of lower uptake. Although larger diameter apertures performed somewhat better for the larger lesions investigated, smaller apertures are recommended for imaging smaller lesions. Overall, larger apertures yielded lower SNRs for smaller lesion sizes, indicating that the enhanced spatial resolution of the smaller pinholes outweighs the loss in sensitivity.

Image reconstruction with maximum likelihood expectation maximization or other statistical techniques may further improve these results because of the inherent reduction of artifacts, such as from streaking or axial detector blurring, as well as the incorporation of modeling the system response that is possible with those techniques. Further observer studies of optimal aperture size via observer models and receiver operator characteristics may also be warranted to evaluate the efficacy of the technique in a clinical lesion detection task.