KAHIBARO
29.1. TOF PET Principles
Table of Contents
Difference in arrival times
In positron emission tomography, each useful event is a coincidence between two 511 keV photons that were created at almost the same time and travel in nearly opposite directions. In conventional PET, you detect these two photons in two different detector elements, record where they interacted, and build a line of response that connects the two detector positions. The coincidence is accepted if the two detections happen within a coincidence time window that is typically several nanoseconds. However, conventional PET ignores the exact difference in arrival times inside that window and treats all accepted coincidences equally along the line of response.
Time of flight PET, usually abbreviated TOF PET, keeps track of the precise detection times of the two photons. For each coincidence event, the scanner measures a detection time $t_1$ in detector 1 and $t_2$ in detector 2. The time difference is defined as
$$
\Delta t = t_1 - t_2.
$$
If the annihilation happens exactly in the middle between the two detectors, the two photons travel the same distance, so they should arrive at the same time, and the expected time difference is $\Delta t \approx 0$. If the annihilation point is closer to detector 1, the photon going to detector 1 travels a shorter distance and arrives earlier. In this case $t_1 < t_2$ and $\Delta t$ is negative. If the annihilation point is closer to detector 2, the situation is reversed, $t_2 < t_1$, and $\Delta t$ is positive.
In an ideal world with perfect timing, the measured $\Delta t$ would be determined only by the true emission position. In reality, the electronics and detector physics introduce timing uncertainty. The timing resolution of a PET system is usually characterized by a full width at half maximum value, often reported in picoseconds. For example, a clinical TOF PET system might have a timing resolution of 300 ps FWHM. This means that if the true time difference between two photons is zero, the measured time differences form a Gaussian distribution centered around zero with a width of about 300 ps at half its maximum height.
The speed of light, which is approximately $c = 3\times 10^8 \,\text{m/s}$, connects time differences to distance differences. A time difference of 1 ns corresponds to a path length difference of about 30 cm. Since two photons start from the same point and move in opposite directions, the spatial uncertainty along the line of response is related to the timing resolution by a factor of one half.
Important relationship between timing resolution and spatial resolution along the LOR:
$$
\Delta x \approx \frac{c\,\Delta t}{2},
$$
where $\Delta x$ is the spatial uncertainty along the line of response and $\Delta t$ is the timing resolution (for example, FWHM) of the coincidence measurement.
As an example, with $\Delta t = 300 \,\text{ps}$, the corresponding spatial uncertainty is
$$
\Delta x \approx \frac{3\times 10^8 \,\text{m/s} \times 300\times 10^{-12}\,\text{s}}{2} \approx 4.5\,\text{cm}.
$$
This means that instead of being completely uncertain along the entire length of the line of response, the annihilation point is likely to lie within a segment of roughly 4.5 cm along that line. The better the timing resolution, the shorter this segment becomes, and the stronger the potential benefit for image quality.
In a GATE simulation, TOF information is represented as detection times in singles and coincidences. The digitizer for a TOF PET system applies timing blurring to the true interaction times in the crystals. The coincidence sorter then uses these blurred times to build coincidences and to compute the measured time differences. When you analyze TOF PET simulation output, you can look at histograms of $\Delta t$ to understand the timing resolution and to check the effect of timing blurring. TOF-aware reconstruction algorithms use these time differences to improve image signal to noise ratio, especially in large patients where random and scattered events are more problematic.
Localization along the LOR
The fundamental idea that makes TOF PET powerful is that the difference in arrival times can be converted into an estimate of where along the line of response the annihilation occurred. If you consider two detectors placed on a line with an annihilation somewhere in between, you can describe the position of the event by a coordinate $x$ measured along the line, with $x = 0$ at the midpoint. If the event is exactly in the middle, the photons travel equal distances and $\Delta t = 0$. If the event is shifted by a distance $x$ toward detector 1, the distance to detector 1 becomes shorter and to detector 2 becomes longer by the same amount.
The distance difference between the two photons is $2x$. Since both photons travel at the speed of light, the difference in their flight times is
$$
\Delta t = \frac{2x}{c}.
$$
Solving this for $x$ gives a direct formula to estimate the position from the measured time difference,
$$
x = \frac{c\,\Delta t}{2}.
$$
TOF position estimate along the line of response:
$$
x_{\text{TOF}} = \frac{c\,\Delta t}{2},
$$
where $x_{\text{TOF}}$ is the estimated displacement from the midpoint between the two detectors, and $\Delta t$ is the measured arrival time difference.
In an ideal noiseless system, this equation would give the exact annihilation position along the line of response. In a real scanner, timing blurring means that the measured $\Delta t$ is only an approximation of the true time difference. This leads to an uncertainty in $x_{\text{TOF}}$, which is directly related to the timing resolution as described earlier. You can think of each coincidence event as contributing probability along the line of response, with the highest probability at $x_{\text{TOF}}$ and decreasing away from that point according to a Gaussian function whose width is determined by the timing resolution.
In image reconstruction, traditional non TOF PET assumes that each coincidence event provides uniform information along the entire line of response. TOF PET instead assigns a spatial weighting along the line that is peaked around the TOF estimated position. This more localized contribution reduces noise propagation in the final image. In particular, TOF PET improves image signal to noise ratio, improves contrast recovery for small lesions, and reduces artifacts due to inaccurate attenuation correction and scatter.
From a simulation perspective, GATE itself does not perform image reconstruction. However, it can provide all necessary TOF information in the output data. Each coincidence record can include the detection times in each detector, so you can compute $\Delta t$ and the corresponding $x_{\text{TOF}}$ for every event. When you process GATE output with ROOT or Python tools, you can:
- Compute the time difference $\Delta t$ between the two detectors in a coincidence.
- Convert $\Delta t$ to a TOF position $x_{\text{TOF}}$ along the line connecting the two detector centers.
- Use this TOF position to create TOF weighted lines of response that are then passed to a dedicated TOF reconstruction algorithm.
In analytical discussions of TOF benefits, an important concept is the TOF kernel width, which is the effective length along the line of response over which an event contributes. The TOF kernel width is closely related to $\Delta x$ from the timing resolution. When the object being imaged has a transverse size much larger than the TOF kernel width, TOF information can significantly reduce image variance. In contrast, if the object is small compared to this width, the benefit is less dramatic.
In practice, the speed of light is very large, so even subnanosecond timing resolutions correspond to several centimeters of spatial uncertainty. As detector technologies improve and TOF resolutions reach 200 ps or even 100 ps, the kernel width shrinks, and localization along the line of response becomes more precise. This trend motivates careful simulation of TOF performance in GATE, where you can vary timing blurring parameters, detector geometry, and physics settings to study how much improvement a given TOF resolution provides for a specific scanner design or clinical task.
Views: 12