*2.4. Hardware*

#### 2.4.1. Silicon Photomultiplier (SiPM) Technology

The photomultiplier converts the photon emitted from the scintillation crystal into an electronic signal. Compared to conventional photomultiplier tubes (PMT), silicon photomultipliers (SiPM) in recently developed PET/CT scanners are smaller, provide up to 100% coverage of the scintillation crystal surface and offer high photon sensitivity, low noise, and fast timing resolution [44]. The chief advantages of SiPM are higher effective sensitivity and improved signal-to-noise ratios (SNR).

To estimate the independent added value of SiPM technology, performance metrics can be compared with otherwise similar PMT-based scanner models. Most importantly, these scanners should feature a similar length of the axial field of view (FOV) because the system

sensitivity (cps/kBq), a main determinant of the SNR, increases quadratically with an increase in axial FOV (e.g., system sensitivity gain of approx. 80% for an axial FOV increase of 33%) [45]. The potential gain associated with SiPM is dependent on the relative coverage of the detector area with SiPM elements and their specific timing resolution relative to the PMT comparator. If, for example, a SiPM system with low relative coverage of <60% and a modest timing resolution of 382 ps [44,45] is compared to a bismuth germanate (BGO) PMT scanner with similar axial FOV length [46], the SiPM system's sensitivity can be similar or even slightly inferior (20.8 vs. 23.3 cps/kBq) [45]. In contrast, a SiPM system with 100% detector coverage and 214 ps timing resolution [47] can offer increased system sensitivity over a comparable PMT system with lutetium oxyorthosilicate (LSO) crystals [48] that exceeds the expected difference based on the slightly larger axial FOV (16.4 vs. 9.6 cps/kBq with a 25.6 vs. 22.1 cm FOV) [47]. If further improvements in timing resolution are achieved with future generations of SiPM [44], this could bring with it a proportional increase in effective sensitivity (noise equivalent counts; NEC) [49,50]. Such a gain in counts could translate into an SNR gain that is equivalent to the square root of the relative improvement in timing resolution [49,51].

Additionally, the development of small SiPM elements has enabled small voxel sizes down to 1 × 1 × 1 mm<sup>3</sup> [52] with potential superiority over PMT-based systems in reconstructed spatial resolution and quantitative accuracy in the case of small lesions [53–56].

## 2.4.2. PET/CT vs. PET/MRI

CT-based attenuation correction (CT-AC) for PET has the advantage of directly and reliably translating tissue densities from CT Hounsfield units (e.g, acquired with 120 keV) into linear attenuation correction factors for 511 keV photons [57]. In contrast, MR-based attenuation correction (MR-AC) in PET/MRI relies on different, vendor-dependent MR sequences that are used to automatically segmen<sup>t</sup> the body into three or four compartments (air, bone, soft tissue, and fat) [58]. These compartments are assigned specific linear attenuation correction factors for 511 keV photons. However, several sources of error exist, including extracorporeal components such as headphones, dense hair or MR coils [59], incorrect attenuation correction factors for bone and lung tissue [60], incorrect segmentation of such compartments, and MR image truncation [61]. Systematic underestimation of SUV in bone tissue or lesions close to the bone of >10% compared to PET/CT has been described [60,62], whereas SUV differences were <5% in other normal tissues [60]. It should also be noted that cross-calibration of PET/MRI scanners is generally hampered by waterfilled phantoms, which are commonly used for SUV calibration in PET/CT and can produce substantial artifacts in the MR-AC sequence with corresponding SUV errors [63,64]. In such phantoms, inter-vendor differences in MR-AC maps [61] can lead to SUV differences in phantom spheres of 10–20% [58].

Studies that have investigated normal organ and lesion SUV in patients undergoing both PET/CT and PET/MRI after a single injection of [18F]FDG were mostly biased by a non-randomized scan order. Lesion SUV were usually lower on the first examination, which can partly be explained by the increasing [18F]FDG accumulation in most (malignant) lesions beyond 60 min after tracer administration. However, systematically different normal organ SUVs have been reported for [18F]FDG as well, which most likely result from MR-AC inaccuracies [60,65,66]. Furthermore, as such studies can only compare one single PET/CT scanner with one PET/MRI scanner and have usually only investigated one specific set of reconstruction parameters, it is difficult to draw a generalized conclusion regarding the comparability of SUV in clinical PET/CT and PET/MRI.
