Clinical PET and PET/CT

• Clinical PET and PET/CT : Quantitation | Image Quality | Protocols | Scanner Geometries

Improving Imaging Capabilities for PET and PET/CT Scanners

Supported by NIH grants R01-CA42593 (Lewellen), R01-CA74135 (Kinahan), P01-CA42045 (Krohn), and a research grant from General Electric Medical Systems (Kinahan/Lewellen).

The UW IRL conducts research synergistically with the he UW Nuclear Medicine clinic for basic, translational, and clinical studies. We provide support for other research activities of the division as well as pursuing related research that is clustered into four areas:

• Improving quantitation for PET and PET/CT scanners
• Development of tools to quantitatively measure image quality
• Optimizing clinical acquisition protocols
• Investigating alternative scanner geometries

Improving quantitation for PET and PET/CT scanners

Our work in Improving quantitation for PET and PET/CT scanners falls into four categories: (1) validation of corrections for isotopes with complex decay schemes, (2) development of improved image reconstruction algorithms, (3) improving the accuracy of inclusion of X-ray CT images into attenuation correction for PET/CT scanners, and (4) incorporation of PET images into radiation treatment planning systems.

Radioimmunotherapy with 131 I and 90 Y labeled monoclonal antibodies require accurate biodistribution measurements prior to therapy for reliable dosimetry. The quantitative advantage provided by PET over single photon imaging has inspired considerable interest in the positron emitting isotopes of iodine ( 124 I) and yttrium ( 86 Y). Unfortunately, these isotopes are characterized by prompt cascading gamma ray emissions and photon energies capable of septal penetration and positron-electron pair production, which contribute unwanted backgrounds to the PET image data; without compensation, these backgrounds can produce unacceptable bias in the final image. Figure 10 depicts some of the results on the complex isotope correction scheme development showing the impact of modeling the additional events due to pair production form the high energy gamma   rays in 86Y. Similar correction schemes are being validated for 124 I and 94 Tc.

Figure 10:   Data Spectrum anthropomorphic torso phantom scanned with 86Y in liver and soft tissue regions.   Spine in phantom contains no activity.   The Standard reconstruction (a) yields erroneous activity in spine that was not recorded in 86Y bias corrected image (b).

For conventional positron-emitting isotopes, we are developing image reconstruction techniques that more accurately model the acquisition physics. Specifically we are looking at the impact of more accurately modeling statistical noise and detector resolution blurring. Figure 11 illustrates the effect of including the effect of attenuation on photon statistics. The effect of modeling detector blurring (for a small anima scanner) was illustrated in figure 12 .

Figure 11. Illustration of the different 3D whole-body PET image reconstruction methods from the same patient data. (a) 3DRP. (b) FORE+AWOSEM. The combination of FORE+AWOSEM leads to improvements in image SNR in clinically feasible reconstruction times. In addition we investigated the comparative performances of the 3D reconstruction algorithms for tumor detection with human observers using a volumetric observer tool that uses the same display as our clinical PET oncology imaging (fig. 4).   These results also indicated an advantage for the FORE+AWOSEM over the 3DRP 3D image reconstruction algorithm.

Figure 12. Image resolution improvements with including detector blurring effects.

The advent of dual modality PET/CT scanners has significantly enhanced the physician's armamentarium for the diagnosis and staging of cancer as well as for therapy planning and monitoring response to therapy. The PET/CT scanner platform has new synergies, primarily the use of the X-ray CT image for attenuation and scatter correction of the PET emission data, as well as the use of the CT image as an anatomical prior for the PET image reconstruction. There are, however, relevant scenarios where improving the quantitative accuracy of PET/CT imaging is both important and challenging due to respiratory motion, partial volume effects, or estimation of the attenuation coefficients for high atomic number materials such as bone, metal, or contrast agents. We hypothesize that accurate quantitation can be achieved through the combination of three approaches: (1) combining low-dose X-ray imaging with dual energy CT solely for PET attenuation correction, (2) use of the CT image as an anatomical prior for the PET image reconstruction, and (3) the use of respiratory-gated low-dose CT for attenuation correction. Figure 13 illustrates the effects of motion and contrast agents on PET/CT(AC), while figure 14 demonstrates the potential improvements from including the CT image as an anatomical prior for the PET image reconstruction.

Figure 13. top: CT scan with i.v. contrast agent. Bottom: PET images are reconstructed with CT-based attenuation correction, with and without i.v. contrast agent. PET difference image is superimposed on CT showing combined effects of motion (white arrows) and contrast (yellow arrows) in attenuation correction induced artifacts.

Figure 14. Transverse sections through a CT image volume and reconstructed PET volume images of a test 3D phantom showing four hot contrast objects and three cold contrast objects showing the effect of including anatomical (CT) information with measured data.

Recent publications related to this project:

Harrison, R., S. Dhavala, P. Kumar, Y. Shao, R. Manjeshwar, T. Lewellen, and F. Jansen. Acceleration of SimSET photon history generation. in IEEE Nuclear Science Symposium and Medical Imaging Conference. 2002. Norfolk, VA: IEEE p. 1835-1838.

Harrison, R., D. Somasekhar, N. Prasanth, Y. Shao, and T. Lewellen. Importance Sampling in PET Collimator Simulations. in IEEE Nuclear Science Symposium and Medical Imaging Conference. 2003. Portland, OR. p. in press.

Lee, K., P. Kinahan , J. Fessler, R. Miyaoka, and T. Lewellen. Pragmatic Image Reconstruction for the MiCES Fully-3D Mouse Imaging PET Scanner. in IEEE Nuclear Science Symposium and Medical Imaging Conference. 2003. Portland, OR.: IEEE p. in press.

Lee, K., P. Kinahan, R. Miyaoka, J. Kim, and T. Lewellen, Impact of system design parameters on image figures of merit for a mouse scanner. IEEE Trans. Nucl. Sci., 2004: p. in press.

[top]

Development of Tools to Quantitatively Measure Image Quality

With the rapidly increasing use of PET for cancer detection and staging, unanswered questions about the impact of the choice of acquisition, processing, and reconstruction parameters on image quality are becoming more important. To quantitatively assess task-dependent measures of image quality relevant to clinical PET oncology imaging, we have developed volumetric analysis techniques for human and model observer studies. We have applied a multi-target approach (to improve the sensitivity of detection studies) using accurately simulated PET data to compare alternative acquisition, processing, and reconstruction strategies. The ranking of different methods by human observers has been analyzed using (1) a simple non-parametric fraction-found metric and (2) the area under the alternate free-response ROC (AFROC) curve for detection SNR.

We have developed and tested (1) methodologies for human observer studies of lesion detection based on the volumetric display software used in practice for clinical PET imaging, (2) model (numerical) observers based on 3D extensions of the channelized Hotelling observer (CHO) and non-prewhitening matched filter (NPW). Our results indicate that volumetric (3D) model observers behave differently than planar (2D) model observers, and appear to correlate better with volumetric human observer studies ( figure 15 ). These were quantitatively analyzed by alternative free-response ROC (AFROC) analysis and non-parametric methods. These methodologies have been successfully applied to investigations of the impact on lesion detection of the effect of: (1) acquisition mode (2) data pre-processing, (3) image reconstruction, and (4) post-preconstruction smoothing.

Figure 15. Average human observer SNR as calculated from an AFROC analysis compared to SNR(CHO) for targets located in the lungs as a function of the target contrast for three different acquisition protocols.

Recent publications related to this project:

Kinahan, P., J. Kim, C. Lartizien, C. Comtat, and T. Lewellen. A Comparison of Planar Versus Volumetric Numerical Observers for Detection Task Performance in Whole-Body PET Imaging. in IEEE Nuclear Science Symposium and Medical Imaging Conference. 2002. Norfolk, VA p. 1267-1271.

Lartizien C, Kinahan PE, and Comtat C, A Lesion Detection Observer Study Comparing 2D Versus Fully-3D Whole-Body PET Imaging Protocols. Journal of Nuclear Medicine, vol. 45, pp. (to appear), 2004.

Kim J-S, Kinahan PE, Lartizien C, Comtat C, and Lewellen TK, A Comparison of Planar Versus Volumetric Numerical Observers for Detection Task Performance in Whole-Body PET Imaging. IEEE Transactions on Nuclear Science, vol. (accepted), 2004.

Cheng PM, Kinahan PE, Comtat C, Kim J-S, Lartizien C, and Lewellen TK, Effect of scan duration on lesion detectability in PET oncology imaging. In: 2004 IEEE International Symposium on Biomedical Imaging, Arlington, VA, April 15-18, 2004. vol. (to appear), 2004.

[top]

Optimizing Clinical Acquisition Protocols

In spite of the rapid growth of PET oncology imaging protocols, there has been no systematic analysis or study performed of the effect of patient scanning protocol on image quality or clinical PET task performance. We have developed a model of the noise effective count rates (NEC) adjusted for injected activity that can be used to choose the dose/weight as a function of body mass index (BMI) to maximize aggregate image quality and also to determine the effect of patient dimensions on aggregate image quality ( figure 16 ).

Figure 16. Model of how the noise effective count rates (NEC) depends on patient morphology and as a function of body mass index (BMI).

Our hypothesis is that the effects of patient morphology scan duration, and tracer uptake levels on image figures of merit can be predicted. We are extending our scanner count rate model to predict how patient morphology, scan duration, and tracer uptake levels affect image quality, as measured by quantitative figures of merit under the assumption that nothing else changes. We are also determining the effect of the reconstruction regularization parameters on image figures of merit.   Our goal is to determine the smallest detectable lesion size (or the detection probability for a range of sizes and tracer accumulation) as a function of the patient morphology, activity at scan start, partial volume corrected tracer uptake, and scan duration. We will also determine if reconstruction parameters that are optimal for numerical observers reduce the minimum size (or range) of detected lesions compared to standard clinical reconstruction parameters. This can clearly effect patient management if, for example, metastases that would otherwise be missed are detected. Finally, these studies will indicate the feasibility of adjusting the activity injected and scan duration to compensate for patient morphology.

Recent publications related to this project:

Beaulieu S, Kinahan PE, Tseng J, Dunnwald LK, Schubert EK, Pham P, Lewellen B, and Mankoff DA, SUV Varies with Time After Injection in 18F-FDG PET of Breast Cancer: Characterization and Method To Adjust for Time Differences. Journal of Nuclear Medicine, vol. 44, pp. 913(abstract), 2003.

Lartizien C, Comtat C, Trebossen R, Kinahan PE, Ferreira N, and Bendriem B, Optimization of the injected dose based on Noise Equivalent Count (NEC) rates for 2D and 3D Whole-Body PET. Journal of Nuclear Medicine, vol. 43, pp. 1268-1278, 2002.

Kinahan PE, Lartizien C, Chander S, Meltzer CC, McCook B, and Torok F, Optimization of Wholebody PET FDG Oncology Scanning Protocols. Journal of Nuclear Medicine, vol. 43, pp. 216P (abstract), 2002.

[top]

Investigating Alternative Scanner Geometries

The scanner geometry optimization project is investigating alternatives for 2D and fully-3D PET systems. The approach is to develop a coarse septal-collimation system that will provide an optimization between sensitivity for trues and the acceptance of scatter and random events. We term this approach '2.5D' and are initially investigating collimator optimization for the GE Advance PET scanner. We will then couple these results with simulations of potential system designs where we keep the total volume of scintillator a constant (typically 11 liters for a modern whole body system) and look at the best distribution of that volume (crystal size, ring diameter, and axial extent of the detector array). To determine the distribution of true, scattered, and random events for all of these configurations, SimSET is being used to for the initial investigation. The SimSET results will be used to develop a parametric model that can be used with our analytical simulator, ASIM, which will then be utilized to determine the impact on image quality. This will require many thousands of images to be generated and analyzed using numerical observers to determine the efficiency of detection.

Recent publications related to this project:

Kohlmyer, S., C. Stearns, P. Kinahan , and T. Lewellen. NEMA NU2-2001 performance results for the GE Advance PET system. in 2002 IEEE Nuclear Science Symposium and Medical Imaging Conference. 2002. Norfolk, VA p. 890-894.

Surti S, Badawi RD, Holdsworth C, El Fakhri G, Kinahan PE, and Karp JS, A Multi-Scanner Evaluation of PET Image   Quality Using Phantom Studies. In: 2003 IEEE Nuclear Science Symposium and Medical Imaging Conference, Portland, OR, October 19 - 25, 2003.

[top]