Evaluation of the MRI Compatibility of PET Detectors Modules for Organ-specific Inserts in a 3T and 7T MRI Scanner

Abstract

Background

Simultaneous positron emission tomography (PET)/magnetic resonance imaging (MRI) scanners and inserts are valuable tools for accurate diagnosis, treatment planning, and monitoring due to their complementary information. However, the integration of a PET system into an MRI scanner presents technical challenges for a distortion-free operation.

Purpose

We aim to develop a PET insert dedicated to breast imaging in combination with the 3T PET/MRI scanner Biograph mMR (Siemens Healthineers) as well as a brain PET insert for the 7T MRI scanner MAGNETOM Terra (Siemens Healthineers). For this development, we selected as a basis the C13500 series PET modules (Hamamatsu Photonics K.K.) as they offer an all-in-one solution with a scalable, modular design for compact integration with state-of-the-art performance. The original PET modules were not designed to be operated with an MRI scanner, therefore we implemented several modifications such as signal transmission via plastic optical fiber, radio frequency (RF) shielding of the front-end electronics, and filter for the power supply lines. In this work, we evaluated the mutual MRI compatibility between the modified PET modules and the 3T and 7T MRI scanner.

Methods

We used a proof-of-concept setup with two detectors to comprehensively evaluate a potential distortion of the performance of the modified PET modules whilst exposing them to a variety of MR sequences up to the peak operation conditions of the Biograph mMR. A method using the periodicity of the sequences to identify distortions of the PET events in the phase of RF pulse transmission was introduced. Vice versa the potential distortion of the Biograph mMR was evaluated by vendor proprietary MRI compatibility test sequences. Afterwards, these studies were extended to the MAGNETOM Terra.

Results

No distortions were introduced by gradient field switching (field strength up to 20 mT/m at a slew rate of 66.0 T/ms⁻¹). However, RF pulse transmission induced a reduction of the single event rate from 33.0 kcounts/s to 32.0 kcounts/s and a degradation of the coincidence resolution time from 251 ps to 299 ps. Further, the proposed method revealed artifacts in the energy and timing histograms. Finally, by using the front-end filters it was possible to prevent any RF pulse induced distortion of event rate, energy, or time stamps even for a 700° flip angle (45.5 μT) sequence.

The evaluations to assess potential distortions of the MRI scanner showed that carefully designed RF shielding boxes for the PET modules were required to prevent distortion of the RF spectra. The increase in B₀ field inhomogeneity of 0.254 ppm and local changes of the B₁ field of 12.5% introduced by the PET modules did not qualitatively affect the MR imaging with a spin echo and MPRAGE sequence for the Biograph mMR and the MAGNETOM Terra, respectively.

Conclusion

In conclusion, our study demonstrates the feasibility of using a modified version of the PET modules in combination with 3T and 7T MRI scanners. Building upon the encouraging MRI compatibility results from our proof-of-concept detectors, we will proceed to develop PET inserts for breast and brain imaging using these modules.

1. INTRODUCTION

Positron emission tomography (PET) and magnetic resonance imaging (MRI) are two medical imaging modalities that are used for diagnosis, treatment planning, and therapy monitoring in fields such as neurology^1,2, oncology³, and cardiology⁴. PET, in comparison to MRI, has a superior potential to detect and characterize cancer based on biochemical and molecular processes. This is enabled by the outstanding high sensitivity of PET in combination with target-specific tracers, which allows for the detection of molecules in the picomolar range.⁵

In contrast to PET, MRI can be used to gather high-resolution anatomic and functional information. The major advantages of MRI are the improved spatial resolution, high soft-tissue contrast, and radiation-free operation, as well the as the variety of techniques, such as diffusion imaging, magnetic resonance spectroscopy, and functional MRI.⁶ To fuse the complementary information provided by the two imaging modalities, combined whole-body PET/MRI scanners were first introduced for preclinical purposes; they were later found to have wide application potential and the first clinical scanner, the Biograph mMR (Siemens Healthineers, Erlangen, Germany) was introduced in 2010^7–9. Instead of whole-body imaging, a different approach is pursued by organ-dedicated MR compatible PET inserts (e.g., for the brain¹⁰ or the breast¹¹). The principal motivation for using a PET insert is the increased image quality and higher detection sensitivity gained in comparison to whole-body scanners. First, better image quality regarding image resolution helps to identify smaller structures, which can, for example, improve lesion detection and enhance the accuracy of tumor classification. Second, the proximity of the PET insert to the organ of interest yields a high sensitivity due to the high solid angle coverage, which can either be used to further improve the image quality or the patient comfort by reducing the scan time.

In particular, we aim to develop a PET insert dedicated to breast imaging in combination with the whole-body 3T PET/MRI scanner Biograph mMR¹² as well as a brain PET insert for the 7T MRI scanner MAGNETOM Terra (Siemens Healthineers, Erlangen, Germany).

We have selected the C13500 series PET modules (Hamamatsu Photonics K.K., Hamamatsu, Japan) as a common basis for both inserts for the following reasons: First, as their application-specific integrated circuit (ASIC) based technology offers state-of-the-art performance^13,14 with compact dimensions suitable for integration into both inserts where space for detector electronics is limited. Second, the ASIC is accompanied by further components such as an on-chip high voltage supply with temperature compensation as well as a back-end system for the processing of the PET event data and communication with a PC (refer to¹⁴ for detailed information on the PET modules). Third, as this all-in-one solution provided by a single manufacturer in combination with a modular and scalable technology enables high flexibility in the design process and prototyping of both inserts. To build both inserts based on the PET modules we have to overcome two inherent limitations of the original PET modules.

On the one hand, the commercially available portfolio of the PET modules targets clinical whole-body PET systems with a scintillation crystal cross area of 3.1×3.1 mm². To extend the modules for high-resolution systems, support for smaller scintillation crystals is required. Therefore we developed a detector based on the PET modules which enables a 1.5×1.5 mm² scintillation crystal cross area using a light sharing approach which proves the suitability of the PET modules for high-resolution PET inserts¹⁵.

On the other hand, major difficulties arise due to the operation of the PET insert inside an MRI scanner^16,17. The high static magnetic field strength can severely affect the PET insert components and in particular, the photosensors, therefore current clinical PET/MRI scanner are based on solid-state photosensors^1,18,19, and the detectors in this work use silicon photomultipliers (SiPMs). Next to the static magnetic field, the PET insert must withstand switching gradients, i.e., fast changes in the local magnetic field strength, as well as transmission of radio frequency (RF) waves. Further, the PET insert should not harm the operation of the MRI scanner, which can be introduced due to electromagnetic interference or magnetic field inhomogeneity caused by magnetic susceptibly differences.^20–22

The original C13500 series PET modules were not designed to fulfill these criteria for mutual MRI compatibility, therefore, in a first step, modifications were implemented in this work including signal transmission via plastic optical fiber, RF shielding of the front-end electronics, and filter for the power supply lines. We then comprehensively evaluated a potential distortion of the performance of the PET modules whilst exposing them to a variety of MR sequences up to the peak operation conditions of the Biograph mMR. Vice versa the potential distortion of the Biograph mMR was evaluated by vendor proprietary MRI compatibility test sequences. Finally, these studies were extended to the MAGNETOM Terra to evaluate the potential MRI compatibility of the modified PET modules not only for 3T (intended for the breast insert) but also for 7T (intended for the brain insert).

2. MATERIALS AND METHODS

2.1. Modified PET Modules for MRI Compatibility

The PET modules used in this work were based on the C13500–4075LC-12 TOF-PET modules (Hamamatsu Photonics K.K., Hamamatsu, Japan). The original back-end electronic components for data acquisition, clock generation, and power supply were used in this work and placed outside of the MRI scanner room to prevent mutual distortion (Figure S-1).

The original PET modules use copper-based transmission lines, which are prone to RF distortion and therefore were replaced in this work with plastic optical fibers and conversion boards from optical to electrical signals to transfer the data and 50 MHz clock signals between front-end and back-end electronics (Figure S-1). Power was provided via shielded cables, shielding braids were connected to the RF shielding cage of the MRI room, and power lines were fed through a low pass filter (Siemens Healthineers, Erlangen, Germany) mounted at the wall outside of the MRI room. A further modification was the replacement of the original high voltage power supply chip C11204–02 on the signal processing board (SPB) including the ASICs (Figure 1) with the C11204–4 chip which does not use an inductor with a ferrite core, prone to distortions by the magnetic field. A customized laser sintered enclosure (Figure 1) was used to house the optical conversion boards, SPBs, and the 12×6 lutetium fine silicate (LFS) scintillator array with 4.14×4.14×20 mm³ crystal size coupled to a 12×6 MPPC matrix series S13360 (Hamamatsu Photonics K.K., Hamamatsu, Japan) with an area and cell size of 4×4mm² and 75 μm. Adhesive copper foils with 18 μm thickness were used to provide RF shielding for the enclosures Figure 2A) and three waveguides were used for the plastic optical fibers and two for cooling with compressed air. In addition to the low pass filter mounted at the wall of the MRI room, a second set of pluggable filter with the same properties (5^th order LC low pass filter with a cut-off frequency of 5.2 kHz) and in a copper shielding enclosure was mounted to the front-end electronics (Figure 2A).

Figure 1:

Two opposing detectors with front-end electronics including SiPM, scintillation crystal block, SPB, conversion board, power cable, and plastic optical fibers. For visualization, the lids of the shielding boxes and shells of DSUB connectors were removed.

Figure 2A:

Shielding boxes housing PET detectors and front-end electronics with mounted front-end filters (inside separate shielding boxes). Figure 2B: Setup centered on patient bed of the Biograph mMR with cylindrical water phantom placed on the top (for visualization position outside the bore, measurements were performed with PET detectors inside the bore).

2.2. Measurement Setup, PET Event Data Acquisition and Analysis in the Biograph mMR 3T

Two opposing scintillator crystal blocks were placed with a distance of 40 mm, to provide a reasonable fast acquisition time, to a centered 353 kBq Na-22 point source (Figure 1, Figure 2A). The detectors were placed on the patient bed (Figure 2B) and positioned in the axial and transaxial center of the 3T PET/MRI scanner (Biograph mMR at University hospital Tuebingen, Germany) for all measurements.

The impact of the front-end filter was evaluated by measurements with and without the filter plugged in. To obtain the MRI signal a cylindrical water phantom was placed on top of the shielding boxes (Figure 2B). Singles event data was recorded for both detectors, stored as binary file, and subsequently processed within MATLAB (MathWorks Inc., Natick, USA). In total, 10⁷ events were recorded for both detectors for each measurement to provide sufficient event statistics for each scintillation crystal.

In order to evaluate the impact of different MR sequences on the PET event data multiple measurements were performed while running different MR sequences (section 2.4.1) and compared against reference measurements for which no MR sequence was applied. Of note, the mounting of the front-end filters required a new alignment of the shielding boxes, to account for the small variation in count rates a second reference measurement was performed with the front-end filters installed.

2.2.1. PET Event Count Rate-related Effects

First, count rate-related effects were evaluated by monitoring the single event count rate for both detectors and the coincidence event count rate within a 120 ns time window. A limitation of monitoring the average count rate over the entire acquisition to evaluate MRI compatibility arises for MR sequences with a low duty cycle (e.g. 3.3% for the rf_pulse sequence with 200° flip angle (FA) in section 2.4.1). Therefore, only a small fraction of the acquired events, which fall in the RF pulse on phase, i.e., time when RF pulses were transmitted, would be potentially distorted and no degradation in the average count rate might be observed.

RF pulses are transmitted in a repetitive pattern, with a fixed pulse width (PW) and repetition time (TR). Therefore, it is possible to extract events in the RF pulse on phase by selecting only PET events with time stamps that fall into these repetitive time windows according to the formula:

$$\text{r}\left(i\right)=\left(Ts\left(i\right)+Sp\left(m\right)\right)\%\text{TR};\text{ r}\left(i\right)\le \text{PW},$$ (1)

where Ts(i) is the time stamp of PET event i and r(i) is the remainder for event i. If the remainder is less than or equal to PW, the event is assigned to subgroup m (see an example for the method to assign PET events into subgroups in Figure S-2). A prerequisite for this is to know the start point Sp of the RF pulse. However, as there was no reference, such as a trigger signal, in the PET event data stream that allowed to identify this start point, events were assigned to subgroups for 100 different start points (to provide a high temporal resolution) according to the formula:

$$Sp\left(m\right)=\frac{TR}{100}*\left(m-1\right)$$ (2)

Then the average number of single events for each detector and subgroup was calculated, which allowed to evaluate a potential distortion of the count rate in the RF pulse on phase.

2.2.2. PET Event Energy Rate-related Effects

Potential distortions of event energies were assessed by obtaining energy histograms for the individual scintillation crystals and a Gaussian fit was used to obtain photopeak position and full-width-at-half-maximum (FWHM) energy resolution. Average energy resolution, photopeak position, and corresponding standard deviation were reported for the two detectors. Of note, the PET modules used a time-over-threshold (ToT) with 4 ns granularity to obtain the energy information, therefore units of energy were in ns. Total block energy histograms of detector 2 were used to qualitatively assess potential artifacts due to distortions induced by MR sequences.

Again, the low duty cycle of 3.3 % for the 200° rf_pulse sequence might only affect a small fraction of the events and thus potential distortions in the total block energy spectrum might not be visible. Therefore, energy spectra in the RF pulse on phase as well as in the RF pulse of phase, i.e., all events that do not fall in the RF pulse on phase, were analyzed for potential artifacts. The starting point to determine the RF on phase for this sequence was set to 10.5 ms which marks the subgroup with the lowest number of events (Figure 3A).

Figure 3A:

Number of single events per subgroup for both detectors, applied 200° FA rf_pulse sequence and without front-end filters; B: with front-end filters.

This method to identify the RF on phase requires identifying a decrease in the number of counts per subgroup. However, as the front-end filters prevented this decrease in count rate the previously mentioned concept to evaluate the events in different subgroups was extended to evaluate 100 different total block energy histograms for the 200° and 700° FA rf_pulse sequence. To visualize these energy histograms a color-coded representation was used, where pixel values represent the number of events in the respective bin of the total block energy histogram. This analysis was performed for a measurement with the 200° FA rf_pulse sequence and without the front-end filter mounted for validation of the method. Further to evaluate if the front-end filters were able to sufficiently suppress RF distortion, the measurements with front-end filters and 200° and 700° FA rf_pulse sequence were analyzed.

2.2.3. PET Event Timing Rate-related Effects

To evaluate event time stamp related effects a total block coincidence timing histogram was obtained for events in an energy window of ±1σ around the photo peak position and for a 120 ns time window. This large time window was necessary to assess potential distortions outside the region of the main Gaussian distribution, which is defined as an interval of ±1 ns around the centroid of a Gaussian fit of the timing histogram. The FWHM of a Gaussian fit was reported as coincidence resolving time (CRT). The ratio of events that have time stamp differences outside the main Gaussian distribution to the total number of coincident events in the 120 ns window was calculated to assess potential distortions of the time stamps due to applied MR sequences.

Analogs to the examination for the energy spectra to overcome the limitations in monitoring distortions for the low duty cycle for the 200° FA rf_pulse sequence, timing histograms in the RF pulse on phase were compared to RF off phase. Further, the previous analysis using color-coded representations for the energy histograms was applied to the timing histograms to determine distortions due to 200° and 700° FA rf_pulse sequences with and without front-end filters mounted.

2.3. Measurement Setup, PET Event Data Acquisition and Analysis in the MAGNETOM Terra 7T

After completing the measurements in the Biograph mMR a similar evaluation was performed in the 7T MRI scanner (MAGNETOM Terra at Athinoula A. Martinos Center, Boston, MA, USA). The Na-22 point source had an activity of 394 kBq, a constant acquisition time of 120 s (~6.9×10⁶ events to provide sufficient event statistics for each scintillation crystal) was used and the MRI signal was obtained by a NaCl water bag placed in the center of the MRI scanner. Of note, for measurements in the 7T scanner no front-end filter was used as the attraction of the ferromagnetic connectors of the filter boards impeded their use. Further, the global performance parameters, energy resolution, photo peak position, single event count rate, coincidence event count rate, and CRT were used to evaluate MRI compatibility for the MR sequences in section 2.5.

2.4. MR Sequences for Evaluation of Mutual MRI Compatibility with the Biograph mMR 3T

The MR test sequences used for this work were proprietary and accessible via the service menu of the Biograph mMR.

2.4.1. Sequences for Evaluation of Distortion of the PET Modules

In order to assess how the RF pulses inherent to MR sequences affect the PET modules the vendor-specific rf_pulse sequence was used. This allowed transmitting a sequence of repetitive RF pulses with a frequency of 123.21 MHz and adjustable repetition time (TR), pulse width (PW), and flip angle (FA). According to a reference measurement, an RF pulse with 180° FA and 1ms PW corresponded to a transmitted B₁ field $B_{1\mathit{\text{trans}}}$ of 11.7μT. Therefore, B_1trans could be determined according to the formula:

$${\text{B}}_{1\text{trans}}=11.7\mu \text{T}*\frac{\text{FA}}{180°}*\frac{1\text{ ms}}{\text{PW}}.$$ (3)

The duty cycle of a rf_pulse sequence was defined as the ratio of PW to TR, i.e., the relative time of RF pulse transmission. Of note, for RF pulses with a high FA, the duty cycle was limited to prevent damage to the RF transmission amplifiers. In order to evaluate the impact of RF pulses with different FA and duty cycles rf_pulse sequences as stated in Table 1 were used. In order to assess how the switching of gradient fields affected the PET modules, the vendor-specific gradfree_pulse sequence was used, which allowed repetitive gradient pulse cycles. For one cycle, the gradient field strength was first ramped to a peak gradient strength of 20 mT/m in 0.3 ms, i.e., a 66.0 T/ms⁻¹ slew rate, which is the peak performance of the Biograph mMR. The peak gradient field strength was applied for 1.0 ms and subsequently ramped down to 0 mT/m in 0.3 ms. In the second half of the gradient pulse cycle, this pattern was repeated with a negative polarity. Switching of the gradients in this manner was performed simultaneously for the x, y, and z-axis gradients.

Table 1:

Settings for different rf_pulse sequences

FA [°]	PW [ms]	TR [ms]	Duty Cycle [%]
30	1.0	3.0	50.0
90	1.0	3.0	33.3
200	1.0	30.0	3.3
700	1.5	100.0	1.0

2.4.2. Sequences for Evaluation of Distortions of the Biograph mMR 3T

To evaluate potential distortions of the RF receiving spectrum of the Biograph mMR the vendor-specific rf_noise sequence was used. The bandwidth of the RF power spectrum was ±500 kHz with a center frequency of 123.21 MHz. The spectra were obtained for measurements with the PET modules and compared against a reference measurement without the PET modules. To show the importance of the shielding boxes for suppression of RF distortion, the RF power spectrum was also acquired with the PET modules in a previous version of the shielding boxes. In this previous version the edges of the shielding boxes were not soldered (Figure 2A), which resulted in gaps for the RF waves to potentially propagate from the PET modules to the receiver coil of the MRI scanner. Next to the qualitative comparison of the RF power spectra, the following metrics were used to quantify potential differences: the noise floor level (obtained by a two-term Gaussian fit of the histogram of the RF power values) and the standard deviation of the RF power values.

The vendor-specific field sequence was used to assess B₀ field homogeneity. An MR image of the cylindrical phantom obtained by this sequence will show interference stripes. Each interference stripe corresponded to a change in the B₀ field by 0.254 ppm. Measurements were performed with the PET modules and compared against a reference measurement without the PET modules.

In the same way, the rf_field sequence was used to evaluate the RF field (B₁) homogeneity. Here each additional interference strip in comparison in comparison to a reference measurement corresponded to a change in the B₁ field of 6.25 %. In order to qualitatively evaluate how images obtained by a commonly used imaging sequence were affected by the PET modules, a spin echo sequence with a TR of 400 ms, FA of 90° FA, 400 ms TR and 12 ms echo time (TE) was used.

2.5. MR Sequences for Evaluation of Mutual MRI Compatibility with the MAGNETOM Terra 7T

In contrast to the measurements for the Biograph mMR, for the studies in the MAGNETOM Terra, no dedicated vendor-specific test sequences were used to evaluate potential distortions of the PET modules. Instead, a more practical approach was followed by selecting different sequences which are widely used for 7T MR imaging, such as a turbo spin echo (TSE), gradient echo (GRE), and echo planar imaging (EPI) sequence. Following this practical approach a magnetization-prepared rapid gradient-echo (MPRAGE) sequence was used to qualitatively assess potential distortions of the PET modules on the operation of the MAGNETOM Terra with a 3D-printed realistic head phantom²³. The phantom consisted of four compartments (bone, brain, muscle, and internal air cavities) based on the MRI scan of a healthy volunteer. The skull was printed as solid plastic and the brain compartment was filled with a gel made of 0.12% NaCl and 3% agar. Of note, the TSE, GRE, and EPI sequences were subsequently applied; however, the measurement with the MPRAGE sequence was performed on another day. As such, an additional reference measurement for the MPRAGE sequence was performed.

To evaluate potential distortions of the RF receiving spectrum of the MAGNETOM Terra the vendor-specific rf_noise_spectrum sequence was used. This sequence used a Fourier transformation of the received RF signals to obtain the RF power spectrum within a receiving band of 1 MHz around the Larmor frequency.

3. RESULTS

3.1. Evaluation of Distortions of the PET Modules from the Biograph mMR 3T

The impact of the rf_pulse and grad_freepulse sequences on various PET performance parameters is shown in Table 2. A more detailed evaluation of count rate, energy, and time stamp related effects is described in the following sections.

Table 2:

Performance parameters of the PET modules for reference measurements and applied MR test sequences for the Biograph mMR 3T. Abbreviations: detector 1,2 (D1,D2); single event count rate per detector (CRSingle); coincidence event count rate (CRCoin).

Test Sequence	Front End Filter	Energy Resolution (D1) [%]	Energy Resolution (D2) [%]	Photo Peak (D1) [ns ToT]	Photo Peak (D2) [ns ToT]	CRSingle (D1) [kcounts/s]	CRSingle (D2) [kcounts/s]	CRCoin [kcounts/s]	CRT [ps]
Reference	no	10.4±0.5	10.2±0.5	434.9±21.8	434.6±24.2	33.0	32.5	1.88	251
rf_pulse, FA=30°	no	10.3±0.5	10.0±0.5	433.5±22.1	433.8±24.4	33.0	32.6	1.87	299
rf_pulse, FA=90°	no	10.4±0.5	10.0±0.5	433.3±22.1	433.6±24.3	32.0	32.6	1.76	265
rf_pulse, FA=200°	no	10.5±0.5	10.1±0.5	434.1±21.9	434.4±24.2	33.0	32.4	1.85	253
grad_freepulse	no	10.4±0.5	10.1±0.5	434.1±22.1	434.0±24.3	32.9	32.6	1.89	255
Reference	yes	10.4±0.5	10.2±0.5	434.3±21.8	433.2±24.4	33.8	34.1	1.97	252
rf_pulse, FA=30°	yes	10.3±0.5	10.05±0.5	433.6±21.8	433.3±24.2	33.8	34.2	1.95	252
rf_pulse, FA=90°	yes	10.4±0.5	10.04±0.5	434.0±22.0	433.0±24.3	33.8	34.1	1.97	254
rf_pulse, FA=200°	yes	10.4±0.5	10.1±0.5	434.1±21.8	433.5±24.3	33.8	34.1	1.96	252
rf_pulse, FA=700°	yes	10.3±0.5	10.0±0.5	434.0±22.0	433.4±24.2	33.8	34.2	1.96	253

3.1.1. PET Event Count Rate-related Effects

Count rates remained stable in comparison to the reference measurement (Table 2) except for the 90° FA rf_pulse sequence, for which a reduction for detector 1 from 33.0 kcounts/s (reference) to 32.0 kcounts/s was observed. This reduction in the count rate could be prevented by the front-end filters for all applied rf_pulse sequences (Table 2).

In order to assess potential distortions that were not reflected by the average count rate due to the low duty cycle of the rf_pulse sequence, the evaluation of the number of events per subgroup was used for the 200° FA rf_pulse sequence (Figure 3). The maximum decrease in the number of events per subgroup was 8.4 % for detector 2 (Figure 3A). A Gaussian fit of this curve resulted in a FWHM of 1.05 ms. As this corresponded to the PW of 1.0 ms for this sequence the RF on phase was identified with a starting point of 10.5 ms which was used for subsequent evaluation of energy and time stamp related effects. Of note, the decrease in number of events per subgroup was larger for detector 2 than for detector 1. This is assumed to be caused by the different routing of the power cable, which for detector 2 had a longer part of the cable inside the field of view, thus being more prone to RF distortion.

The decrease in number of events per subgroup could be prevented by the use of the front-end filters and low variation was obtained over the different subgroups, 1.51×10⁵±356 and 1.52×10⁵±347 events for detector 1 and 2, respectively. Further with the front-end filters also the number of events within subgroups remained stable for the 700° FA rf_pulse sequence, 4.53×10⁴±229 and 4.59×10⁴±208 for detector 1 and 2, respectively.

3.1.2. PET Event Energy-related Effects

Energy resolution and photo peak position remained stable for all sequences in comparison to the reference measurement (Table 2) even without the front-end filters. The total block energy spectrum shows a small deviation for the 90° FA rf_pulse sequence (Figure 4A), which could be eliminated by the front-end filters (Figure 4B). The energy histogram for the events in the RF on phase of the 200° FA rf_pulse sequence showed artifacts when no front-end filters were used (Figure 5A, B). The mean energy of the events was reduced to 394.1 ns (RF on phase) in comparison to 458.7 ns (RF off phase).

Figure 4A:

Total block energy histograms for detector 2 and applied rf_pulse sequences without front-end filters; B: with front-end filters.

Figure 5A:

Total block energy histograms during the RF pulse off and on phases for detector 2 and applied 200° FA rf_pulse sequence without front-end filters; B: energy range from 100 ns to 500 ns.

Without knowledge of the exact starting point of the RF pulse sequence, the proposed method which involved total block energy spectra analysis for different subgroups and a color-coded representation showed the distortion of the energy histograms (Figure 6A). The artifact in Figure 5 translated to the horizontal strip in Figure 6A. Further, the number of events per subgroup that were in an energy window from 100 ns to 380 ns was calculated and the progression over the different subgroups showed a Gaussian distribution with a FWHM of 1.06 ms (Figure 6B). As this corresponded to the PW of 1.0 ms for this sequence, the method was validated to evaluate potential distortions of the event energies in the RF on phase. The color-coded energy histogram representations for the 200° and 700° FA rf_pulse sequences with front-end filters showed no distortions (Figure 6C, D).

Figure 6A:

Color-coded representation of energy histograms for 100 subgroups, applied 200° FA rf_pulse sequence and without front-end filters; B: Number of events with an energy between 100 ns and 380 ns for each subgroup. C: 200° FA rf_pulse sequence with front-end filters; D: 700° FA rf_pulse sequence with front-end filters. 14

3.1.3. PET Event Time Stamp-related Effects

The only test sequences that caused a degradation of the 251 ps reference CRT, were the rf_pulse sequences with 30° and 90° FA and a CRT of 299 ps and 265 ps, respectively (Table 2, Figure 7A). A lower number of coincident events fell into the main Gaussian distribution of the timing histogram for the 30°, 90°, and 200° FA rf_pulse sequence, indicated by the lower amplitude of the respective timing histograms (Figure 7A). The logarithmic representation of the timing histogram for the total 120 ns time window (Figure 7B) showed the increase of events with a time stamp difference outside the main Gaussian distribution for these sequences. The ratio of events outside the main Gaussian distribution to the total number of coincident events was 2.9 % (reference), 2.7 % (grad_freepulse), and 3.7 %, 21.7 %, and 4.4 % for the 30°, 90° and 200° FA rf_pulse sequences, respectively. The use of the front-end filters resulted in distortion-free timing histograms (Figure 7C, D) and event ratios of 2.7 % (reference), 2.9 %, 2.9 %, 2.8 %, and 3.0 % for the 30°, 90°, 200°, and 700° FA rf_pulse sequences, respectively.

Figure 7A:

Timing histograms for applied rf_pulse and grad_freepulse sequences without front-end filters; B: Same timing histograms as in (A) but with logarithmic scale and 120ns time window; C: Timing histogram for applied rf_pulse sequences with front-end filters; D: Same timing histograms as in (C) but with logarithmic scale and 120 ns time window.

The timing histogram for the events in the RF on phase showed a reduction of the probability for events to fall into the main Gaussian distribution (Figure 8A) and the CRT was 269 ps in comparison to 254 ps for the RF pulse off phase when no front-end filters were used. The probability for events with a time stamp difference outside the main Gaussian distribution increased for the RF on phase, and this sequence introduced spurious peaks in the timing histogram in the RF on phase (Figure 8B)

Figure 8A:

Timing histograms during the RF pulse off and on phases and applied 200° FA rf_pulse sequence and without front-end filters; B: Same timing histograms as in (A) but with a logarithmic scale and 120ns time window.

The method used in 3.1.2 for the total block energy spectra to evaluate potential distortions of event energies in the RF on phase with multiple subgroups and a color-coded representation was used in the following to assess time stamp distortions (Figure 9A). Further, the number of events per subgroup that had time stamp differences between −50 ns and −2 ns were calculated and the progression over the different subgroups showed a Gaussian distribution with an FWHM of 1.09 ms (Figure 9B). As this corresponded to the PW of 1.0 ms for this sequence, the method was validated to evaluate potential distortions of the time stamps in the RF on phase. The color-coded timing histogram representations for the 200° and 700° FA rf_pulse sequences with front-end filters showed no distortions (Figure 9C, D).

Figure 9A:

Color-coded representation of timing histograms for 100 subgroups, applied 200° FA rf_pulse sequence and without front-end filters; B: Number of events with a time stamp difference between −50 ns and −2 ns for each subgroup; C: 200° FA rf_pulse sequence with front-end filter; D: 700° FA rf_pulse sequence with front-end filters.

3.2. Evaluation of Distortions of the PET Modules from the MAGNETOM Terra 7T

The impact of the MR sequences for the MAGNETOM Terra 7T on various PET performance parameters is shown in Table 3. Count rates, energy resolution, and photo peak position remained stable in comparison to the reference measurement. The 252 ps CRT (reference) was maintained for the EPI sequence and the other MR sequences caused a minor degradation to 268 ps (TSE), 261 ps (GRE), and 261 ps (MPRAGE).

Table 3:

Performance parameters of the PET modules for reference measurements and applied MR sequences for the MAGNETOM Terra 7T. Abbreviations: detector 1,2 (D1,D2); single event count rate per detector (CRSingle); coincidence event count rate (CRCoin).

Test Sequence	Front End Filter	Energy Resolution (D1) [%]	Energy Resolution (D2) [%]	Photo Peak (D1) [ns ToT]	Photo Peak (D2) [ns ToT]	CRSingle (D1) [kcounts/s]	CRSingle (D2) [kcounts/s]	CRCoin [kcounts/s]	CRT [ps]
Reference	no	10.6±0.6	10.4±0.5	445.3±22.0	441.8±25.4	23.7	26.8	1.05	252
TSE	no	10.6±0.6	10.4±0.6	444.5±21.9	441.6±25.4	23.6	27.0	1.03	268
GRE	no	10.6±0.5	10.3±0.6	444.1±21.9	441.8±25.5	23.7	27.2	1.03	261
EPI	no	10.6±0.6	10.3±0.5	446.3±22.0	443.7±25.4	23.6	27.3	1.01	252
Reference	no	10.4±0.6	10.1±0.5	455.6±22.3	451.9±23.9	24.9	27.5	0.96	252
MPRAGE	no	10.3±0.5	10.0±0.5	453.2±22.7	448.8±24.2	25.5	29.6	0.94	261

3.3. Evaluation of Distortions of the Biograph mMR 3T from the PET Modules

3.3.1. Distortion of the RF Receiver System of the MRI Scanner

The RF power spectra for the previous non-soldered configuration of the shielding boxes demonstrated distortions in the form of spurious peaks (Figure 10A). This distortion could be eliminated with the current version of the shielding boxes which resulted in an RF power spectrum (Figure 10B) comparable to the reference measurement (Figure 10C). These observations correspond to the quantitative evaluation of the noise floor levels and standard deviation of 46.6±6.7, 45.9±1.6, and 46.3±1.6 for the non-soldered shielding boxes, soldered shielding boxes, and the reference measurement, respectively.

Figure 10:

RF power spectra obtained by the rf_noise sequence and red line marking the noise floor. A: PET modules inside previous non-soldered version of the shielding boxes; Figure 10B: current soldered version of the shielding boxes; C: reference measurement without the PET modules.

3.3.2. B₀, B₁ Field Homogeneity and Image Quality

The measurement with the field sequence revealed an additional interference stripe introduced by the PET modules (Figure 11A), which corresponded to an increase in the B₀ field inhomogeneity of 0.254 ppm. The rf_field sequence showed two additional interference stripes for the measurement with the PET modules (Figure 11B), which corresponded to a change in the B₁ field by 12.5 % in comparison to the reference measurement. A qualitative comparison of the images obtained with the spin echo sequence showed that no distortions or artifacts were caused by the PET modules (Figure 11C).

Figure 11:

Comparison of images for the reference and measurement with the PET modules obtained by A: the field sequence to assess B0 field homogeneity; B: by the rf_field sequence to assess B1 field homogeneity; C: the spin echo sequence to assess image quality.

3.4. Evaluation of Distortions of the MAGNETOM Terra 7T from the PET Modules

The RF receiving spectrum showed no distortions, e.g., spurious peaks, when the PET modules were operated inside the MRI scanner (Figure 12A). This indicated that the shielding boxes mitigated the RF distortion introduced by the PET modules to a level that allowed the normal operation of the MRI scanner. In accordance with this, the image of the head phantom (Figure 12B) showed no visible distortions or artifacts caused by the PET modules.

Figure 12A:

RF receiving spectrum acquired with the rf_noise_spectrum sequence (average of 30 single spectra shown in red); B: Image of the head phantom with the MPRAGE sequence.

4. DISCUSSION

Within this work, we evaluated whether the Hamamatsu PET modules are suitable to be used for inserts that are operated within an MRI scanner. We first had to implement several modifications to the original PET modules to enable a potential distortion-free operation. The challenge of a natural incompatibility between PET insert and MRI scanner was present for the first solid-state photosensor-based brain insert¹, as well as for more recently developed brain inserts^24–27 or a small ring structure²⁸. Despite the different designs and MRI scanners used for these PET inserts all of them share the concept to avoid mutual distortions between both systems. The first consideration is to minimize the components of the PET insert, which are exposed to harsh conditions inside the MRI scanner. For this reason, in this work, we located the back-end electronics of the PET modules outside of the MRI scanner room. Second, in general, PET inserts use a type of data transmission, such as differential, coaxial, or optical, that has a high insensitivity to RF interference. This was realized in this work via an optical fiber transmission for data and shielded cables for the power supply were used. Third, the components of the PET insert that must be located inside the MRI scanner are always encapsulated in shielding enclosures of materials such as carbon fiber or thin layers of copper, as used in this work, which mitigate RF distortion of the PET insert by the MRI scanner and vice versa. Fourth, to mitigate the remaining distortions that could not be encountered by the previous actions, filters commonly located at the wall of the MRI scanner room are used as in this work (together with an additional filter for the front-end electronics).

In this work, we tested for mutual MRI compatibility, i.e., potential distortions of the MRI scanner caused by the PET modules and vice versa. To evaluate the impact on the modified PET modules first was examined to which extent PET performance parameters were affected by different MR test sequences with the Biograph mMR 3T scanner in comparison to a reference measurement without MR sequence. It was shown that distortions were not introduced by gradient field switching (field strength up to 20 mT/m at a slew rate of 66.0 T/ms⁻¹), however different kinds of distortions were caused by the RF pulse transmission. Such an RF distortion induced reduction of the single event rate from 33.0 kcounts/s to 32.0 kcounts/s and reduction of the 251 ps CRT to 299 ps could both be eliminated once the front-end filters were used. Further with the front-end filters it was possible to eliminate distortions that were visible in the total event timing and energy histograms for the measurements without the filters.

The sequences with a high FA had a low duty cycle, therefore only a small portion of the total number of events could potentially be affected by RF distortion. Therefore, PET performance parameters and energy and timing histograms for the total set of events were limited metrics to ensure no distortions were introduced by these sequences. Therefore, a method using the periodicity of the test sequences was introduced and validated which allowed to explicitly investigate effects on the events in the RF on phase, i.e., during the time of RF pulse transmission. This method revealed a reduction in the RF on phase of the mean event energy from 458.7 ns to 394.1 ns as well as artifacts in the respective energy and timing histograms. With this method, it was also possible to show that even for a 700° FA sequence with a low duty cycle of 1.0 %, the front-end filters prevented any RF pulse induced distortion of event rate, energy, or time stamps.

From the comprehensive evaluation studies with the Biograph mMR, potential distortions of the PET modules could be identified to origin from RF pulses and could be successfully eliminated with the front-end filters. For the MAGNETOM Terra 7T scanner, we used typical imaging sequences, which were a composition of RF pulse transmission and gradient switching, to test for potential distortions of the PET performance parameters. The only impact reported was a small degradation of the CRT, e.g., 16 ps for the TSE sequence. From the previous comprehensive evaluation, we conclude that this degradation was caused by the RF pulses inherent to these sequences, which could be eliminated in the future by front-end filters.

The evaluations in this work to assess potential distortions of the MRI scanner caused by the PET modules showed that carefully designed RF shielding boxes for the PET modules were required to prevent distortion of the RF spectra. Further, to achieve MRI compatibility, it was necessary to preserve the homogeneity of the B₀ and B₁ fields²⁰. The reported increase in B₀ field inhomogeneity of 0.254 ppm and local changes of the B₁ field of 12.5% were acceptable for the Biograph mMR according to the vendor guidelines to assess MRI compatibility of third-party products. It is assumed that the main source for B₀ homogeneity degradation in this work was the nickel-coated connectors of the power lines, which should be replaced while designing inserts based on the PET modules. Furthermore, shimming can be used to account for potential low local variations in the B₀ field.

The degradation of the B₁ field homogeneity is assumed to be introduced by the copper-based shielding boxes. RF fields were expected to be absorbed at their surfaces with subsequent secondary RF field generation causing an attenuation of the primary RF field distribution²⁹. Therefore, the choice of the RF shielding material, thickness, and assembly needs to be carefully considered for the design of the inserts. Further, this study showed that the B₀, B₁ inhomogeneity and RF distortion introduced by the PET modules did not qualitatively affect the MR imaging with a spin echo and MPRAGE sequence for the Biograph mMR 3T and the MAGNETOM Terra 7T, respectively.

We are aware that more comprehensive tests such as B₀, and B₁ maps and a quantitative signal-to-noise-based evaluation of the MR images would be necessary to attest a full MRI compatibility of the PET modules as they were performed by other groups^30–32. We intend to perform these tests for both inserts, which may show results different from this study due to the higher number of PET modules, changes in geometric design and RF shielding. In addition, as part of the design process, we intend to determine whether the copper shielding used in this work is suitable for the inserts. Although this has been shown to be an appropriate shielding approach for inserts^6,28, it could be beneficial to reduce the thickness of the shielding in combination with a segmented structure. Another approach reported to provide good shielding is to mesh the conductive surface to disrupt the current loop path, e.g., with phosphor bronze metal wires³³. This helps to reduce gradient induced eddy currents, which can cause secondary field generation, vibration and heating.

Of note, even with the solid copper layer shielding in this work, we did not observe any heating induced effects such as a shift in the photopeak position or intense heating of the shielding boxes. To assess temperature changes, we performed measurements with a demanding series of sequences in terms of eddy currents (typical MPRAGE, EPI, TSE, and diffusion sequences, as well as isolated XYZ gradient pulsing (+/+ polarity, 45 mT/m, 90 T/m/s, 75% duty cycle)) and according to color-changing indicator labels, the surface temperatures of the signal processing board, SiPM, and enclosure shielding did not exceed 40°C. A measurement outside the MR scanner showed a maximum temperature of 38°C, so the air-based cooling provides sufficient cooling for this setup. However, we expect increased heating due to the more compact design of the inserts. It has been reported that air-based cooling^10,35, as used in this work may be sufficient, but if necessary, we will consider a combination with thermal pads and liquid cooling³⁶.

This work aimed to show that the PET modules together with our modifications are suitable for the development of PET inserts in a 3T and 7T MRI scanner. These inserts would require the transition from the bulky proof-of-concept detector prototype we used in this work towards a more compact design suitable to be integrated on a larger scale into the inserts. Therefore, we currently employ together with Hamamatsu Photonics K.K. a new version of the conversion board which is capable to provide a data transmission for multiple PET modules.

5. CONCLUSION

With this study, we aimed to assess whether the Hamamatsu PET modules can potentially be used to build a PET insert that can be operated in a 3T and 7T MRI scanner. For this purpose, two detectors based on a modified version of the PET modules were used in combination with a 3T and 7T MRI scanner to evaluate mutual MRI compatibility, i.e., potential distortions of the MRI scanner by the detectors and vice versa. We have shown that distortions of the PET events caused by the RF transmission of the MRI scanner could be eliminated by using RF shielding boxes and front-end filter. We further demonstrated that B₀ and B₁ homogeneity could be preserved such that simultaneous operation of the PET modules and MR imaging was possible. Based on the encouraging results regarding MRI compatibility obtained with the proof-of-concept detectors in this work, we will develop two PET inserts for breast and brain imaging based on the PET modules.

Supplementary Material

Fig S2

Figure S-2: Visualization of the method for assigning PET events according to their time stamp Ts to subgroups based on the repetitive pattern of rf_pulse sequences. Stencils with a width PW and repeating after TR are shown for exemplary starting points Sp(m=1) and Sp(m=2). In this example, the events i= 1,4 are assigned to subgroup m=1 and events i=2,4,7 are assigned to subgroup m= 2.

Fig S1

Figure S-1: Front-end of the PET detector with conversion boards used for optical signal transmission via plastic optical fibers to the back-end electronics. Block diagram for visualization of the main components, their location and connections.