Additional Gains with Time-of-Flight PET at High Counting Rates: Lessons Learned from Early Time-of-Flight PET Systems

TO THE EDITOR: Recent publications on the new generation of time-of-flight (TOF) PET cameras by Lois at al (1), Surti et al. (2), and Karp et al. (3) very elegantly show the reduction in image noise and improvement in lesion detection with TOF PET positron cameras using phantoms and clinical studies. These new TOF PET cameras are optimized for high-resolution tumor detection using ¹⁸F-FDG as the tracer and show that the highest improvement in signal-to-noise ratio in the image is obtained when imaging large 35- to 40-cm-diameter objects. Their noise reduction results are consistent with the published result by Yamamoto et al. (4) in 1982 for 35-cm-diameter objects scanned at low counting rates using the early Super PETT I TOF PET camera built at Washington University, Saint Louis, Missouri. However, Yamamoto et al. (4) also showed that, at high counting rates, there is an additional gain with TOF PET due to the way random coincidences are treated with TOF PET. Therefore, there is the possibility of even better performance capabilities with TOF PET when imaging at higher counting rates.

The early TOF PET systems (4–6) were designed for fast imaging with short-lived isotopes. Experience with these systems showed that there are 3 areas of PET in which the TOF information obtained with fast detectors can significantly improve image quality over non-TOF PET. The first is the low-counting-rate mode of imaging whereby only the true counts are taken into consideration and the improvement in image signal-to-noise ratio is expressed as the square root of D/d, where D is the diameter of the object and d is the TOF resolution measured as the full width at half maximum in centimeters. The mathematics of TOF PET signal-to-noise ratio is elegantly described in a teaching editorial by Budinger (7). The gains obtained by TOF PET were simulated by Wong et al. (8) for different-sized objects and different TOF resolutions. Basically, in lay terms, the use of TOF information helps place the detected positron closer to the real location of the radioactivity during the image reconstruction process, thus improving the image quality. And the larger the object, the greater the improvement.

The second area in which TOF PET can improve image quality is in the reduction of random coincidences during dynamic imaging at high counting rates. When one is imaging organs such as the brain and the heart with short-lived isotopes such as ¹⁵O, ¹¹C, ¹³N, and ⁸²Rb, random coincidences can degrade image quality. Using the Super PETT I, Yamamoto et al. (4) measured the signal-to-noise gain at higher counting rates to be as high as 2.8, compared with 1.7 at lower counting rates. This experimental observation was subsequently confirmed by Holmes et al. (9) using a mathematic model of the effects of accidental coincidences in TOF PET systems.

The third area of image quality improvement in TOF PET systems is in the reduction of dead-time losses in the system at high counting rates. The data acquired during a typical scan comprise true counts, scattered counts, and random counts. Randoms increase as the square of the radioactivity in the field of view, whereas true and scattered counts increase linearly with the activity. Therefore, at higher counting rates, randoms can be higher than true counts and compete with true counts for transfer to the computer. As randoms increase at higher counting rates, true counts are disproportionately reduced because of dead-time losses in the system. The combination of randoms and dead-time losses can reduce the noise equivalent counts, the equivalent of good counts, for large objects such as the abdomen at higher counting rates as shown by Surti et al. (2). Their data show that randoms and scatter counts are equal to true counts at a much lower counting rate for a 35-cm-diameter object than for a 20-cm-diameter object. Thus, any increase in injected dose to the patient beyond this point will yield reduced improvement in signal-to-noise ratio in the image while increasing the radiation dose to the patient.

Since the introduction of the early TOF PET systems in the 1980s, detectors, electronics, and computer performance have increased dramatically. Things that were not possible at that time can be implemented in the present TOF PET systems to further enhance the signal-to-noise ratio by reducing the effects of randoms and dead-time losses. One of these advances is the ability to dynamically change the coincidence acceptance window to fit the object being imaged. The coincidence timing windows in the early systems were fixed and controlled by the position of the attenuation ring at the periphery of the field of view. Today, it is possible to use the attenuation data to selectively shorten coincidence timing acceptance windows during data acquisition for different parts of the body. As an example, for brain imaging, the coincidence window can be reduced to accept only counts emitted from the brain by using the TOF information, thus reducing the randoms collected and reducing the dead-time losses for the system. So, although the paper by Lois et al. (1) shows a smaller improvement with TOF PET in imaging the head than in imaging the abdomen at lower counting rates, it may be possible to increase the signal-to-noise ratio when imaging the brain at higher counting rates with ¹⁵O or ¹¹C. Even if only ¹⁸F-FDG is used to do the imaging, the possibility of obtaining first-pass blood flow in tumors with ¹⁸F-FDG during the first 2 min after injection of the ¹⁸F-FDG bolus will result in significantly higher randoms and dead-time losses. Therefore, the anticipated gain with TOF PET in the future may be higher than published by Lois et al. (1) and Surti et al. (2) at higher counting rates as the new generations of TOF PET cameras are optimized to reduce randoms and dead-time losses.

One area of improvement that I have wished to pursue is selective-area data acquisition and reconstruction. From the first demonstration of TOF PET using a heart phantom (10), we showed that the TOF confidence-weighted back-projected image (without image reconstruction) clearly delineates the heart from the rest of the phantom. This finding suggests that, using the TOF information, it is possible to isolate the heart from the chest during data acquisition. If we can use the TOF information to limit the data acquisition to just over the heart, we may be able to selectively scan a smaller area in a larger object and reconstruct it with an even greater improvement in signal-to-noise ratio. So, even though the static mode of TOF PET did not show a significant gain by Lois et al. (1) in the heart, selective imaging of the heart in the future with ⁸²Rb, ¹¹C, or ¹³N may produce even better results by using the full potential of TOF PET.

This is an exciting time for PET. The development of new and faster detectors has created the possibility that clinical imaging with PET will be faster, have higher resolution, and have better time-of-flight timing than currently. Reduction of TOF timing to one half of the current timing is now feasible, thus opening up new areas of research and development in clinical image improvement, reduction of radiation to patients and smarter imaging with PET. And in the future, as the timing resolution improves even further, we may be able to image selective areas to reduce the effects of randoms and dead-time losses and further enhance image quality.

• © 2010 by Society of Nuclear Medicine