Clinical image quality perception and its relation to NECR measurements in PET

Patient population

A total of 71 consecutive patients (75 exams) were evaluated in this retrospective study. All patients were referred for a clinical FDG-PET/CT from January to December 2012 and underwent a PET/CT-MRI using a tri-modality setup. Based on the retrospective nature of the study, no formal institutional ethics committee approval is needed and the IRB was waived.

Exclusion criteria were uncontrolled glucose levels (higher than 120 mg/dL) and patients who have not fasted for at least 4 h, unwillingness to undergo the additional MR examination, claustrophobia and MR-incompatible medical devices (e.g. cardiac pacemaker, insulin pump, neurostimulator, cochlear implant). Another exclusion criterion was the presence of artefacts in at least one bed position on PET images (six exams in four patients were excluded), which made the clinical reading not applicable and could represent a confounding factor on IQ assessment.

Overall, 75 exams were performed in 71 patients (overall 655 bed positions; 41 males, 30 females; median age of 57 years, range 16 to 89 years). Of those, 59.2% (n = 42) presented with pathologic lesions (primary tumour, lymph nodes, metastasis and/or loco-regional recurrence), while 40.8% (n = 29) did not have any pathologic lesion. The indications for PET/CT were tumour primary staging in 22.5% (n = 16), follow-up or re-staging in 74.6% (n = 53) and 2.8% (n = 2) for non-oncologic reasons (fever of unknown origin and aortic aneurysm graft evaluation). Thirteen (18.3%) patients were under treatment during the exam.

PET/CT imaging

PET/CT imaging was performed on a PET/CT-MR setup (including a time-of-flight Discovery 690 PET/CT and a Discovery 750w 3T MR; both from GE Healthcare, Waukesha, WI, USA).

PET/CT was performed according to the European Association of Nuclear Medicine (EANM) procedure guidelines for tumour PET imaging. Patients fasted for at least 4 h prior to injection of 18F-fluorodeoxyglucose (FDG). The mean FDG injected dose was 311.7 MBq (SD = 21.3, range 231.1 to 373.4). The mean FDG injected dose/body weight was 4.3 MBq/kg (SD = 0.9, range 2.6 to 6.6). Based on the additional MRI examination prior to the PET/CT, the mean delay between injection and PET scan was 95.7 min (SD = 12.7, range 69.9 to 128.3). Unenhanced low-dose CT and PET emission data were acquired from the mid-thigh to the vertex of the skull. CT data were acquired in shallow breathing with dose modulation between 15 and 80 mA, 120 kVp and a pitch of 0.984:1, reconstructed to images of 0.98-mm transverse pixel size and 3.75-mm slice thickness. PET data was acquired in 3D time-of-flight mode with a scan duration of 2 min per bed position, an axial FOV of 153 mm and a 23% overlap of bed positions, resulting in a total PET acquisition time ranging from 16 to 20 min. PET images were formed using a fully 3D TOF OS-EM iterative reconstruction including point-spread-function compensation (VuePoint-FXS) with 3 iterations and 18 subsets onto a 256 × 256 image grid (2.73 × 2.73 × 3.27 mm voxels) over a 700-mm-diameter FOV. Images were filtered in image space using a 4.0-mm FWHM in-plane Gaussian filter followed by an axial filter with a 3-slice kernel using relative weights of 1:4:1. All quantitative corrections, including normalisation, deadtime, randoms, scatter and attenuation, were applied during image reconstruction [7].

Image processing and analysis

The acquired PET and CT images were transmitted to a dedicated review workstation (Advantage Workstation, GE Healthcare), which enables the review of the PET and CT images side by side or in fused/overlay mode (PET/CT).

In a first step, all data were automatically analysed using Matlab (MathWorks, Natick, MA, USA) to estimate the NECR versus activity concentration in a similar manner to the National Electrical Manufacturers Association (NEMA) analysis and based on previous publications [8]-[12]. This measurement is essentially the same usually performed on phantom data, but the required data (the number of true, scattered and random counts) are estimations either provided by the scanner software or automatically extracted from the reconstructed image. To determine the NECR for a bed position of patient data, the following steps were used: the PET raw data files were used to measure prompts and to calculate randoms from singles and obtain the bed position duration. Then, the reconstructed PET images were used to obtain an estimate of the model-based scatter fraction. With those two items, the NECR was calculated for each bed position, and the patient NECR was an average of all whole-body bed positions. By using the PET images, voxels above a fixed activity concentration of 1.0 kBq/mL, the average activity concentration for all voxels in the bed position was calculated. Decay to each bed position based upon its start time relative to scan start was applied as well.

While the measurement of NECR is a quantitative parameter, more clinical and qualitative evaluations were performed in a second step. Here, all the PET/CT exams were read qualitatively by a board-certified nuclear medicine physician/radiologist and a radiologist with substantial experience in PET/CT image reading.

Two subjective scores were defined to evaluate image quality. The IQ local score (IQ_L) was a 3-point scale assigned to each bed position in transversal planes in all patients (n = 655 bed positions), where 1 means poor, 2 means good and 3 means excellent IQ. The IQ global score (IQ_G) was a 10-point scale assigned after evaluating the coronal whole-body PET, where 1 was the worst IQ and 10 was the best. The readers assessed the image quality subjectively based on prevalence of noise, contrast between different tissues and organs (e.g. differentiation between the lung and thoracic wall) and lesion detectability. There was no dedicated evaluation of each of those parameters independently (noise, contrast and detectability).

The advantage of performing both scores remains on the possibility to evaluate the whole-body at one time (IQ global score) and in a more detailed approach, identifying better the relation between the different anatomical structures within the body compartments (IQ local score).

Patient data were analysed concerning overall weight, BMI, FDG dose at the start of acquisition (D _Acq), presence of bowel uptake and presence of FDG-positive pathologic lesions (significantly higher uptake than the surrounding background activity).

For the qualitative evaluation of focal bowel uptake and pathologic FDG-positive lesions, only the bed stations, which showed those lesions, were analysed (e.g. only the two abdominal bed stations were analysed concerning focal bowel uptake and were correlated with the IQ local score).

Two additional parameters were defined for each patient: the ratio between the dose at the start of acquisition (D _Acq) and patient weight (R _DW) and the ratio between D _Acq and patient BMI (R _DBMI). The R _DW and R _DBMI thresholds were determined, above which the resulting image quality was at least good (IQ_L > 2) in more than 90% of patients. The rationale behind these parameters was to create a threshold for different weight/BMI groups with which we could determine the dose needed to achieve high IQ scores throughout our patient group.

Our population was divided into four different groups according to their BMI: group I (BMI lower than 20), group II (BMI between 20.1 and 25), group III (BMI between 25.1 and 30) and group IV (BMI higher than 30.1). BMI groups were analysed concerning the differences in NECR and IQ_L.

Statistical analysis

All statistical tests were performed using SPSS Statistics Version 21 (IBM, Armonk, NY, USA). p values <0.05 were considered statistically significant. Pearson's correlation analysis was performed to evaluate the relation between NECR, IQ_L, IQ_G, weight, BMI, R _DW and R _DBMI. One-way ANOVA test was performed to calculate the difference between the four different BMI groups. Tukey's test was additionally applied to evaluate the differences between each BMI group. Mann-Whitney test was applied to analyse the difference of NECR and IQ_L in patients with bowel uptake and pathologic lesions.

Image quality

In this study, the methods used for the evaluation of IQ were found to be reliable for two main reasons. First, because both the IQ_L and IQ_G, although independent methods, were significantly and strongly correlated to each other (r = 0.94). The second reason was the significant and positive correlation between IQ_L and IQ_G to NECR, a well-established and quantitative way of assessing IQ of PET systems [3],[5].

There were no images scored below 5 on IQ_G or below 1.44 on IQ_L, because all the patients were referred to a clinical PET evaluation and, therefore, required an image with a diagnostic quality.

NECR

NECR measurements represent a method of approximation to image quality in PET systems, at least to a certain extent which seems clinically reliable. However, getting the NECR values is a challenge. Several studies have used NECR based on phantom or simulation measurements [9],[11],[12]. One study has measured NECR in a patient population [5]; however, they compared this value only with image noise (a different automatic technical measurement). One major difference compared to our study is that our estimation of NECR is patient-based and that an additional correlation was made to visual IQ assessments. Thus, we actually have a clinical image estimation, which can be transferred to other PET/CT systems with different detector technologies to compare and estimate the expected clinical IQ. NECR measurements are already widely accepted and used as means of comparing data acquisition performance between different PET systems. By establishing a link between our clinical (subjective) evaluation and the corresponding raw data quality in terms of NECR, it is indeed possible to estimate the tracer doses required to achieve similar results in other scanners. This can be achieved by comparing the phantom-based NECR curves provided by the respective manufacturers.

Correlation to clinical perception

Both IQ_L and IQ_G, as well as NECR, showed a significant positive correlation with the R _DW and R _DBMI.

These results are partially in accordance with the literature. Everaert et al. showed that administration of activities of FDG of ≥8 MBq/kg results in PET images of good to excellent quality in the vast majority of patients using a standardised protocol [4]. On the other hand, Chang and co-workers have demonstrated that an increase in injected dose means a (scanner-dependent) raise in NECR, but only until it reaches the known plateau. From this point on, there is no difference on NECR by increasing FDG injected dose anymore [5].

Our study showed an optimal threshold of 2.6 MBq/kg (R _DW) in which the images are equal or above good IQ. The threshold we found is lower than that published in the literature, because we used the FDG dose during the acquisition of PET (after the uptake time). That holds the advantage that it is not influenced by the delay between injection and PET exam. Using a standardised uptake time of 60 min, the calculated threshold in our setup value would be 3.8 MBq/kg. However, this is still lower than that published in other studies [9]. Since IQ, NECR and R _DW are well correlated with each other, the threshold of R _DW below the clinical image quality is suboptimal and can be translated into a corresponding NECR threshold, which then identifies such suboptimal image quality in a quantitative way.

The EANM recommends standardised FDG activity. For systems with a high count rate capability, the administered FDG activity and scan duration for each bed position should be a relation between the injected dose/kg and time per bed position. Therefore, one may decide to apply a higher activity and reduce the duration of the scan or use reduced activity and increase scan duration - depending on the clinical situation [13].

Putting our technical parameters and clinical evaluations in accordance with those recommendations, the FDG activity would be 6.9 MBq/kg. Our study has shown that PET imaging quality remains good to excellent in the majority of patients even using significantly lower dose.

Weight and BMI

Our study has demonstrated that weight and BMI are negatively and significantly correlated to IQ and NECR. This is expected since other studies have already established this relation. By increasing the patient's weight and BMI, the image quality deteriorates and NECR decreases significantly. This is explainable by the standardised fixed dose of 350 MBq (±10%) that was used in our department at the time period when this study was performed.

As described before, we have also found that in patients with higher BMI, higher FDG doses are needed in order to increase NECR and, hence, image quality. However, it has been shown that once the NECR curve is approaching its plateau it is not useful to increase the injected dose [5]. Furthermore, there are maximal FDG activities defined by the national laws that can be applied [13]. Thus, in current clinical systems (e.g. lutetium oxyorthosilicate (LSO)), FDG activities higher than 530 MBq should not be administered [14]. To increase the image quality in patients where the maximal FDG dose is already given, it is recommended to increase the time per bed position instead of giving higher doses [13].

Our results concerning perceptual and technical differences between weight/BMI groups are also in accordance with the current literature. Recently, Chang and co-workers have demonstrated statistical differences in NECR and signal-to-noise ratio between two different BMI groups [15]. In our population, there was no statistical difference between the BMI groups II and III regarding NECR, while concerning IQ_L, there was no statistical difference between the groups III and IV. One possible explanation for this might be the different appreciation of the image quality between automatic and subjective criteria. In this case, it would appear that the subjective perception of image quality loss with larger patients tends to ‘saturate’ as BMI increases.

Presence of FDG bowel uptake

In clinical routine, partly very high FDG uptake can be seen in the gastrointestinal tract, e.g. in the distal oesophagus, stomach, small intestine and large intestine representing normal patterns of tracer distribution (due to diabetes medication) or inflammatory disease [16]. Diffuse increased FDG uptake in these areas is defined as physiologic and unrelated to the malignant process with relatively high certainty [17]. However, especially medication-related high bowel uptake in patients with diabetes can sometimes impair reading in PET/CT clinical routine.

Our study has found that the presence of bowel uptake is related with (significantly) lower IQ scores, but this relation could not be established for NECR measurements. Thus, although high bowel uptake subjectively impairs the clinical reading, those differences are not reflected within the quantitative measurements.

Our study has shown that the presence of focal FDG uptake, malignant or not, is related to high IQ scores and high NECRs. The reason for the first might be the subjective analysis as well. The presence of FDG focal uptake increases the soft tissue contrast between the structures, enhancing therefore the visual IQ.

Limitations

Our study has several technical limitations. Our data shows a partial correlation between NECR and IQ on average, however, with partially high dispersion. This dispersion of IQ values for a given value of NECR is probably related to the subjective nature of the IQ itself. Although we already performed consensus reading, a given subjectivity cannot be excluded.

Relatively large variations in IQ occur for particular values of NECR. However, the intent of the manuscript was to show a clinically usable approximation between NECR and image quality. This is also why the transferability of our results is not of a quantitative but rather qualitative nature, which is thought to be clinically usable and useful.