Background
Positron emission tomography with computed tomography (PET-CT) is a well-established and fast-growing imaging modality, mainly used in oncology [
1]. Over the years, improvements in detector design as well as the implementation of time-of-flight technology and better reconstruction methods have significantly improved image quality and introduced the possibility of reducing the amount of activity administered and/or scanning time [
2]. A new generation of PET-CT scanners has recently been introduced, and they use silicon (Si) photomultiplier (PM)-based technology. This technology has the potential to increase the detection of pathology, primarily through greater sensitivity. It is hoped that this improved detectability will enable earlier detection of pathologies, including metastatic spread. Thus far, this remains to be confirmed in patient studies. Using the National Electrical Manufacturers Association (NEMA) NU-22012 standard [
3], Hsu et al. [
4] found that SiPM-based PET cameras have greater sensitivity and time resolution compared to conventional PM-based PET cameras. They also presented a clinical case where metastases were detected using the SiPM-based PET camera but not using a conventional PM-based PET camera. Two other studies comparing SiPM-based PET-CT to conventional PET-CT found the SiPM-based PET-CT performed better and detected a greater number of lesions [
5,
6].
The aim of this study was to compare the quality and diagnostic performance of images obtained from a novel SiPM-based PET system (Discovery MI; GE Healthcare, Milwaukee, WI, USA), using the Bayesian penalized-likelihood reconstruction algorithm (Q.Clear), with those obtained from a conventional PM-based PET system with time-of-flight (Gemini TF; Philips Healthcare, Cleveland, OH, USA), using the line-of-response row-action maximum-likelihood algorithm, in phantom and patients undergoing clinical PET-CT scanning with 18F-FDG. Thus, images from two PET-CT cameras with reconstruction algorithms per vendor’s recommendations were compared; i.e. clinically relevant protocols were evaluated.
Discussion
The primary finding of this study was that metabolic TNM classifications indicated tumours were at a worse stage (presence of primary tumour, spread to lymph nodes or distant metastases) for images obtained from the Discovery MI compared to those obtained from the Gemini TF in 27% of patients. Also, a trend toward a greater number of malignant, inflammatory and unclear lesions was found with images obtained with the Discovery MI compared to the Gemini TF (although not statistically significant). Also, lesion-to-blood-pool SUV ratios were significantly greater for Discovery MI images compared to Gemini TF images for lesions smaller than 1 cm. Phantom studies showed better image quality for the Discovery MI.
It is clear from analyses of patient lesions and from the phantom analysis that lesions, especially smaller ones, have higher SUVs and are more visible on images obtained from the Discovery MI compared to those from the Gemini TF. However, SUV measurements can be influenced by many factors, including accumulation time [
8,
9], patient characteristics and reconstruction method [
7,
10‐
12]. Although SUVs were higher in all but one lesion for images from the Discovery MI, a significantly greater lesion-to-blood-pool SUV ratio was found for lesions smaller than 1 cm. Therefore, the greater SUV is likely explained by the greater sensitivity and different reconstruction algorithm of the Discovery MI system rather than longer accumulation times. The new reconstruction algorithm used with the Discovery MI will decrease the effect of partial volume, particularly for small lesions, due to the point spread function in the Bayesian penalized-likelihood reconstruction algorithm. Using the Gemini TF, this partial volume correction was not possible, leading to a more pronounced underestimation of SUVs in small lesions. Also, the smaller voxel size used by the Discovery MI (2.7 × 2.7 × 2.8 mm) compared to the Gemini TF (4 × 4 × 4 mm) contributes to increased SUVs found using the Discovery MI [
13]. A different SUV could also be obtained when using the Q. Clear algorithm because the noise-penalizing determining factor (beta) affects SUV. Thus, our findings depend on both hardware (different generation of PET-CT cameras) and software (different reconstruction methods). The aim of this study was to compare the clinical performances of the two systems. Therefore, the clinical protocols recommended by the vendors were used rather than the same reconstruction method for both.
The greater sensitivity of the Discovery MI permitted shorter times per bed position (1.5 min compared to 2 min for the Gemini TF) with preserved SNRs in the liver for a given dose (4 MBq/kg). The greater axial field of view (20 cm) and the short overlap (24%) permit faster image acquisition. Generally, our experience is that the total PET acquisition time is half that for the Gemini TF. A short acquisition time potentially reduces problems with patient motion and bladder filling and would allow more patients to be examined per day.
It is hoped that the greater sensitivity of SiPM PET-CT and improved reconstruction algorithms will lead to increased detection of pathology, such as earlier detection of small metastases. We found that 27% of patients had different (worse) metabolic TNM classifications using images acquired from the Discovery MI compared to those acquired from the Gemini TF. Unfortunately, we do not know the patients’ true TNM classifications, but our results indicate that more primary tumours, lymph nodes or distant metastases can be detected using the new generation of PET-CT scanners. This agrees with our findings that more malignant and inflammatory lesions were detected (although the difference was not statistically significant, probably because of the limited number of patients included). Also, a greater number of uncertain lesions were found using the Discovery MI PET-CT system. This could be because the physicians interpreting the images were more familiar with images from the Gemini TF, but it could also be a result of the greater number of hypermetabolic lesions found using the novel PET systems. In subsequent studies, where a gold standard is available, it is important to assess both sensitivity and specificity for various diseases using the new generation of PET-CT.
Previous studies of SiPM-based PET platforms in patients have been published. Hsu et al. [
4] used the same SiPM-based PET platform as in our study. They presented performance studies of the Discovery MI system and also included a patient case, comparing with a previous-generation PET-CT. The patient, who had melanoma, was scanned using a Discovery 690 PET-CT system (GE Healthcare) which was immediately followed by imaging using the Discovery MI. Several lesions visible only in the Discovery MI images were found, but no gold standard was used to assess true pathology. Nguyen et al. [
6] compared the diagnostic performance of a SiPM-based PET prototype scanner with a conventional PM PET scanner using 21 patients who underwent clinical
18F-FDG PET-CT. Use of a Gemini TF was followed by use of the SiPM-based prototype, showing better overall image quality with the latter. Lesion SUV
max and lesion-to-blood-pool SUV ratios were significantly greater using the SiPM PET compared to conventional PET, and more so for lesions smaller than 1.5 cm. Baratto et al. [
5] scanned 50 patients using a conventional PM-based PET-CT followed by a SiPM-based PET-CT, finding that using the latter, more lesions were found as well as higher values for SUV
max, lesion-to-blood-pool SUV ratio and liver SUV ratio. The findings of these studies agree with our findings. In all these studies, images were acquired using the conventional PM-based PET-CT first. In our study, we attempted to eliminate differences in accumulation time on the two camera systems by scanning about half the patients first using the Gemini TF PET-CT and half in the reverse order.
Additional studies are needed to establish the full potential value of the new generation of PET-CT scanners, but the results from our study, as well as previous studies, indicate that the increased lesion uptake in the new generation of PET-CT, thanks to improved hardware and reconstruction algorithms, can improve diagnostic performance. It is hoped that smaller lesions can be detected and that this will increase sensitivity and specificity for diagnosing various diseases, potentially improving patient management and outcomes.
Limitations
This study should be viewed in light of some limitations. First, a limited number of patients were included. Second, no gold standard, such as biopsy, or follow-up were performed so that the true number of lesions or true TNM classifications could be assessed. Third, some patients received a diagnostic CT (with intravenous contrast) for one of the PET-CT examinations, whereas others received a low-dose CT. The presence of intravenous contrast is known to slightly affect SUV [
14]. Fourth, because the group who received CTs with the Gemini TF first was larger than the group receiving CTs with the Discovery MI first (by two patients), a significantly shorter accumulation time was seen in images obtained from the Gemini TF, which affects SUV and possibly image interpretation. Fifth, no correction for respiratory movement was applied.
Conclusions
In conclusion, TNM classification was worse in 4 of the 15 patients evaluated for TNM classification, and a trend toward more malignant, inflammatory and unclear lesions was found with images acquired with the Discovery MI compared to the Gemini TF. Also, lesion-to-blood-pool SUV ratios were significantly higher for the Discovery MI compared to the Gemini TF for lesions smaller than 1 cm. It was possible to use a shorter acquisition time for the Discovery MI with preserved SNRs in the liver. Better image quality was found in phantom. Although no gold standard was available, results indicate that the new generation of PET-CT scanners might provide better diagnostic performance, though further studies are needed.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.