JPNM Physics

Visibility of Simulated Lesions Project

Document created May 29, 1996


There are three major ways to image 511 keV photons from positron annihilation:

  1. Conventional PET scanner
  2. Coincidence opposing view, dual-head gamma cameras
  3. Collimated SPECT

It is obvious that collimated SPECT will suffer in lesion detectability due to its combination of poor spatial resolution and poor sensitivity relative to the other two modalities. The object of this study is to define the imaging limitations of collimated SPECT and coincidence gamma cameras relative to that of the conventional PET scanner.

The experiment, which is the visibility of hot lesions in a varying background (i.e., varying contrast), has been performed on four 511 keV-collimated SPECT systems. Preliminary results from the first two gamma cameras were reported at the April 17-19 meeting in Groningen, The Netherlands entitled "The European Conference on PET in Oncology". The presentation is described in the following slides:

Determination of the Visibility of Simulated Lesions in Background with 511 keV SPECT

presented by Douglas J. Wagenaar, Ph.D, Harvard Medical School, on April 19, 1996 in Groningen, The Netherlands.


Slide 1. Title slide

Slide 2. Acknowledgements - Collaborators. This experiment was done on the Elscint Helix at Beth Israel Hospital and the Siemens MS/2 at Dana-Farber Cancer Institute. F-18 was obtained from Massachusetts General HOspital, and perception expertice was offered by Brigham and Women's Hospital's PF Judy, Ph.D.

Slide 3. Acknowledgements - Support. In this slide I acknowledge my present employer for providing the facilities and resources to work on this project. I also acknowledge and disclose Elscint's support in funding my travel expenses to the meeting.

Slide 4. Introduction. Self explanatory. What I mean by "Imaging capabilities" is a measurement of the contrast-detail-count density curve for two collimated gamma camera systems.

Slide 5. Outline of presentation.

Slide 6. We begin our discussion of the project motivation by recapping the nice review of the topic of the various positron imaging methods by Dr. Heikki Minn. (I added this slide after the meeting, since Dr. Minn presented this work at the meeting).

Slide 7. Minn took part in the Michigan study which showed that collimated SPECT tends to miss lesions less than 2.0 cm in extent. He also acknowledged that the Vanderbilt group missed significantly fewer lesions than they did in Michigan. (We don't mention it here, but the conclusions of this study are that the Elscint is significantly better at small lesion visibility that the Siemens. Perhaps that is why Vanderbilt saw more of the smaller lesions?)

Slide 8. Minn's conlucsions about 511 keV SPECT. Pros and cons are self explanatory.

Slide 9. Summary of positron imaging in 1996. Three basis modalities: 1) collimated SPECT; 2) coincidence gamma cameras; 3) BGO PET.

Slide 10. Evolution of Gamma Cameras. The point of this slide is to note that all manufacturers have evolved the gamma camera design to be able to carry substantially more weight than previous gantry designs. This allows optimal collimator designs to be employed. Also, the presenter should mention that in the late 60's - early 70's, rectilinear scanning was done with F-18 fluoride ion in water for bone scanning and a hefty collmator could be employed without the need for a strong gantry.

Slide 11."I may be fat, but I'm ugly." This slide summarizes the dismal outlook for those attempting to do 511 keV SPECT. Not only is the camera "fat" - that is, it has a poor spatial resolution compared with coincidence PET, but it's "ugly" - that is, it will have a deficiency of counts due to poor sensitivity. This latter attribute will result in longer scan times which tend to allow patient motion to occur, making the images blurry. At this point the point of collimated 511 keV SPECT should be driven home: one can do dual isotope or dual energy window scanning with a collimator and one cannot do this in coincidence mode. Furthermore, coincidence gamma cameras cannot stand a high count rate and solution to this problem requires additional research and development. The point of this presentation is to define the imaging capabilities of this poor resolution, poor sensitivity method - the "fat and ugly" method.

Slide 12.Here we define what the experiment is all about - it is a phantom study. In this phantom study we measure sphere visibility with the following variables: varying size of the sphere (there are six spheres), varying contrast (to be explained on a following slide), and varying counts (this is where the Helix evolving images is used to greatest advantage).

Slide 13.Varying size. We used six spheres filled with radioactivity - ranging from 1.0 cm to 3.5 cm in inner diameter.

Slide 14.Varying contrast. This slide explains the entire idea of the experiment. The idea is that the spheres are filled with long-lived (270.8 d) Ge68, while the background is filled with F-18 in an activity concentration at or above that of the spheres.

Slide 15.Ge68 filled spheres. This slide discusses how we obtained the Ge68 filled spheres.

Slide 16.Were the spheres filled correctly? If the spheres were filled correctly and the spheres are the correct dimension, then a plot of activity versus volume should be a straight line. This graph indicates that the smallest spheres were slightly under the straight line curve, most likely due to filling defects - the volume of the fill opening is significant with respect to the volume of the sphere, and this could be reflected in the underfilling of the small spheres. The effect is small, i.e., less than 5%.

Slide 17.Varying counts per scan. Two methods are described. Method one involves using a cycle based on the halflife of F-18: 45, 27, 15, 8, 4, and 2 minute runs were acquired during the course of one 110 minute halflife. When one returns for the second cycle, the contrast has been raised by a factor of two. It turns out that during the long runs, one wastes a lot of contrast measurements, espcecially at the beginning of the experiment when the spheres emerge into visibility from low contrast. It is better to employ method 2: take two minute runs and add them to give varying counts per scan. For example, to get a 46 minute run, sum 23 straight two minute runs. In this way one can finely sample the contrast space (sum runs 1-23, sum runs 2-24, etc.) while getting a potentially large sampling of count-density space (all one has to do is sum different run-lengths of 2-minute runs, for example 8 minutes can be formed from 4 2 minute runs, etc.

Slide 18. Introduction to Evolving Images. State what it really is intended for. Explain that it is a perfect fit for this particular application.

Slide 19. How one actually employs Evolving Images - self explanatory text on slide.

Slide 20.Acquisition parameters. We tried to set up the acquisition for the two gamma cameras to be as nearly the same as possible. The pixel sizes were close, in the neighborhood of 6 mm. The zoom factor was about 1.5 and the matrix size was 64x64. The limitations of disk size prevented a smaller pixel size from being used. Furthermore, this pixel size is closer to clinical relevance. A circular orbit as close to the phantom and support bed was used. The radii listed are the gantry readings. The head-head separations have not yet been compared at these gantry readings.

Slide 21. Definition of terms. Before we go on to how the data were analyzed and interpreted, we must define the following terms: contrast, visibility, and "counts" in the image.

Slide 22. Contrast introduction. This list, provided by Stephen C. Moore, Ph.d., shows how many different ways one might define contrast. We define contrast in terms of the photon emission difference between the sphere volume and the surrounding bath volume.

Slide 23. We define the contrast mathematically to be the signal height (signal minus background) divided by the average of the signal and the background. Here signal and background are the photon emission rates of the spheres and the F-18 bath, respectively. This method, rather than taking a simple ratio of the photon emission rates, avoids an "infinite contrast" when the F-18 has decayed away. An infinite ratio gives a contrast of 2.0 in this equation.

Slide 24. With this definition of contrast, the simple ratios translate into this table of values.

Slide 25. Definition of "visibility". In actually performing these observations, there was little doubt whether a lesion was visible or not, especially since we knew the location and the size. Perceptionists argue that the more pure method is to use a mathematical matched-filter calculation. The Elscint data, 113 consecutive two minute runs, allows one to observe the spheres emerge from the background one by one until even the 1.0 cm sphere is visible. What I found remarkable is that given a spatial resolution in excess of 15 cm at the distances we were operating the system, we could readily resolve the contrast needed between a 1.3 cm and a 1.0 cm lesion!

Slide 26. Definition of "counts" in a scan. We know that the more counts we acquire, the better the image quality and the better the lesion visibility. We needed to quantify the number of counts in a scan, and choose the following method: knowing the time per view, the number of views, the activity of the spheres per cubic centimeter, and correcting for annihilation photon abundance per decay, we calculated how many 511 keV photons emerged from a cubic centimeter of the sphere material. This allows for direct comparison with other methods of positron imaging.

Slide 27. Reconstructed images. This slide shows the beginning and ending 6 mm slice through the center of the sphere plane in the 113 consecutive Elscint runs. In the beginning, no spheres are visible, and at the end all the spheres are visible on the Xpert viewing screen, although in this reproduction it may be difficult to make out the 6th sphere. It must be emphasized that the grey scale manipulation had a far greater impact on the sphere visibility than the type of filter or the cut-off frequency.

Slide 28. Contast-detail-count density. The results of the experiment are summarized in the form of a contrast-detail-count density plot. This plot displays the THRESHOLD contrast necessary to render a certain size lesion visible. Those contrast-size combinations below and to the left of the threshold are invisible; thos contrast-size combinations above and to the right of the threshold are visible. The count density is displayed parametrically; when one has more counts, the threshold moves down and to the left, allowing smaller lesions or lower contrast lesions to be visible.

Slide 29. Construction of the constrast-detail curve. Using method 1 of data acquisition, we have 5 cycles for each count density - more than 9 hours of data. Shown here is one count density's five cycles in the ordinate and the 6 sphere sizes in the abscissa. Using method 2 of data acquisition, the "evolving images" with 2 minute acquisitions, there are 113 contrasts in the ordinate. The threshold curve is sharply defined using method 2.

Slide 30. Elscint results. Shown here are two of the count density threshold curves for the Helix - for 100 M photons per cc and 400 M photons per cc, corresponding to two minute runs and eight minute runs. It is interesting to note that only a 30% signal increase is needed to visualize large lesions. At these count densities, simple contrast ratios of about 6:1 are needed to render a 1.0 cm lesion visilbe. Higher count densities are possible by simple adding more consecutive runs together - for example, a 24 minute acquisition could be reconstructed from 12 summed runs, and this would correspond to a 1200 M count curve. These data are very time consuming to produce due to the slow "ROLLBACK" command. (next slide).

Slide 31.The Rollback command. This very powerful command allows one to select any combination of "evolving images" data and sum them prior to applying reconstruction. Planning is required to run this command on our SP computer, however, since it requires free hard disk space and lots of computation time. Therefore, patient data must be cleared from the disk and the SP must be scheduled for use after clinical hours. The program requires double the disk space of the massive evolving images files. When one runs the program on a set of 40 consecutive spins, it reads each page of the raw projection data back into the display memory as it steps through to create the summed projection set. This takes about 5 minutes for a 15 Mbyte, 40 spin file.

Slide 32. The Siemens MultiSPECT 2 data. Method 1 was used and therefore the time periods during which contrast was low, that is F-18 concentration was high, were eaten up by the long acquisitions of 45, 27, and 15 minutes. Therefore, most of the measurements were taken at high contrast. These curves are considerably above those of the Helix. This means that a 1.0 cm lesion requires fewer counts to be rendered visible on the Helix than on the MS/2. We discuss possible explanations for this difference in the following slides.

Slide 33. Collimator specifications. The collimator specifications do not differe significantly, although the sensitivity of the Siemens collimator is substantially lower than the Elscint as well as all other collimators made by vendors with similar camera heads. Perhaps this sensitivity/specification mismatch is responsible for the poorer performance of this system.

Slide 34.Other possible explanations for the difference in visibility performance include, in addition to collimator specifications, overall detector leadage and electronic performance. One possibility is that the energy window was not correctly set. The Siemens data were taken on two days separated by about three weeks. The first day the camera was "peaked" for 511 keV and the energy spectrum was observed. These data were mostly high contrast since the F-18 concentration was significantly lower than the Ge68 concentration, even at the beginning of the experiment. The second day the acquisition software said the camera was "peaked" at 511 keV and the spectrum was NOT visually checked. I suspect the peak was still in the energy window, however, since the data from the first and the second days overlap and there are not discontinuities in the data.

Slide 35.Conclusions. We have measured the contrast-detail curves for two systems. [since this talk we have measured the collimated ADAC and Park cameras but have not analyzed the data as of June 1, 1996] We now know the "ballpark" contrast and count density needed to visualize a 1.0 cm lesion, something we did not know when we started.

For example, on the Helix, we need a contrast of 1.4 or a simple contrast ratio of 6.4 : 1 and 400 M photons emitted per cc in order to see a 1.0 cm lesion.

After completing the four collimated SPECT systems, we will turn to the coincidence gamma camera MCD of ADAC and finally the BGO PET system at the Massachusetts General Hospital. These data should then clearly demonstrate the performance differences between these three positron imaging methods.

Slide 36. A sentiment many of us without cyclotrons feel.


Douglas J. Wagenaar, Ph..D., wagenaar@nucmed.bih.harvard.edu