The Impact of Patient’s Body Habitus on PET Image Quality in Digital and Analogic PET/CT

Background: New digital versus analogic PET has higher temporal resolution and more stable count rate, potentially limiting the degradation of PET image quality in larger patients. We wanted to describe the inuence of patient’s body habitus on [ 18 F]FDG PET image quality primary in digital PET/CT and analogic PET/CT. Results: We studied retrospectively the relation between patient’s weight, BMI, fatty mass and PET image quality, described by the coecient of variance in the liver (CV liv ) and visually. 177 unique patient exams on digital PET/CT (weight 35-127 kg; BMI 15-44 kg/m 2 ) were performed with 2 protocols (protocol 1: N=52: 3MBq (0,08mCi)/kg [ 18 F]FDG; 2minutes/bed position; 2iterations10subsets; 2mm diameter voxels and protocol 2: N=125: 4MBq (0,11mCi) /kg [ 18 F]FDG; 1min/bed position; 4iterations4subsets; 2mm voxels). 74 unique patient exams were analyzed on analogic PET/CT (weight 38-130 kg; BMI 14-52 kg/m 2 ; with one protocol: 4MBq (0,11mCi)/kg [ 18 F]FDG; 2min40sec/bedposition for BMI<25 and 3min40sec for BMI ≥ 25; 3iterations21subsets; 4mm voxels).


Background
Digital positron emission tomography/ computed tomography (dPET/CT) cameras with silicon photomultipliers (SiPM) detectors are a technological leap and can provide a diagnostic improvement [1]. They show high PET resolution and stable performance in count rate [2]. This is expected to decrease the deterioration of PET image quality (IQ) in function of patient's body habitus.
A comparison of the rst digital time-of-ight (TF) PET/CT and analogic TF PET/CT in the same patients (consecutive imaging with randomized order) have shown a better or equivalent visual PET IQ on the digital camera [3,4].
PET IQ decreases with increasing body mass [5,6] and body mass index [7,8] on an analogic PET/CT (aPET/CT) camera. Signal to noise ratio in the liver (SNR liv ) or inversely its coe cient of variance (CV liv ), and noise equivalent count ratio (NECR) are used as (semi) quantitative indicators of PET IQ. Controversy exists about a linear [7], quadratical [5] or exponential [6,8] relation of these PET IQ parameters with body habitus on analogic PET/CT systems. The liver coe cient of variance (CV liv ) can increase more than 26% between patients weighing less versus at least 70kg in a Japanese study [9] . The European association of nuclear medicine (EANM) imaging guidelines propose two possibilities for calculating the injected activity of [ 18 F] uorodeoxyglucose ([ 18 F]FDG) in function of patient's body mass in a linear and quadratical way [10] .
To overcome degradation of analogic PET IQ in overweight and obese patients we can increase acquisition time and/or injected activity [8,13,15,16]. Also technical progress especially TF and optimization in reconstruction parameters [9,14,17] can contribute to obtain a more constant PET IQ. However little has been described about the necessity and modalities of these compensatory techniques in dPET/CT. Positive in uence of acquisition time on PET IQ in dPET/CT has been reported in 58 oncologic patients [18] Our primary goal was characterizing on digital PET/CT the relation between patient's body mass index (BMI), weight, fatty mass (FM) and [ 18  acquisition with arms along the body; paravenous injection in the image or mentioned in the report; at least one hepatic metastasis; a diffuse or extensive visualized pulmonary or hepatic pathology like evident hepatic steatosis on CT (Houns eld units liver <40); advanced multimetastatic, high uptake disease; recent chemotherapy and growth factors (<2weeks) with important activity shift towards spleen and/or bone and important diffuse bowel, brown fat or muscle uptake.Image protocol Both PET centres are accredited by EANM research limited (EARL) [19] and EANM imaging guidelines were respected [10].
Patients fasted for at least 6h before [ 18 F]FDG injection. Patients' weight was checked on a calibrated scale [20].
In Protocol 2, implemented in a period of increased patient number scanning, we adopted 4MBq (0,11mCi)/kg Image Quality Analysis 1. Semi-Quantitative Analysis-1 spherical volume of interest (VOI) in the liver with a 2,5 to 3cm radius (Nvoxels >>100) was drawn on PSF PET reconstruction data, avoiding upper parts with respiratory artefacts, main large hepatic vessels and tissue bounderies, using Slicer: https://www.slicer.org/ [21].-CV liv was measured in each liver VOI with CV with the * standardized uptake value (SUV) based on the body weight (SUV bw or simpli ed SUV) Measures were automatically extracted with in house software based on ITK [22]. -Patients' weight (kg) and length (m) were extracted from PET DICOM data and veri ed in the PET report -BMI (kg/m 2 ) = weight/ (length) 2 [23] and fatty mass (FM) = weight (kg) -lean body mass (LBM, kg). LBM was estimated with the Janma formula, more adapted for very obese women [24].2. Visual analysis A 5-point global image quality score (Likert score) (visual IQ G ) was given individually by 2 nuclear medicine physicians (with >10 years experience), blinded for clinical information, for 3MBq (0,08mCi) protocol 1 on dPET/CT and aPET/CT. Scoring was de ned as: 5=excellent, 4=good, 3=fair, 2=poor, 1=bad. This global score was based on liver homogeneity, global image noise, image contrast and correct visualization of regions with very low activity (mostly evaluated at intervertebral spaces), as well as the absence of artefacts around high activity regions (bladder). A 10-point score (0-5 with 0,5 interval) of hepatic homogeneity (IQ H ) only was also collected separately on dPET/CT 3MBq (0,08mCi) protocol 1, validated by repeated pairwise comparison in order to classify them [25]. Statistical analysisSTATA version 15 was used for analyses. The normality distribution of BMI, weight and fatty mass were evaluated using a Shapiro-Wilk test. We performed linear regression analysis to investigate the association between weight, BMI, fatty mass and CV liv on each camera. We ran camera dependent strati ed multivariable analyses, and tested whether adjustment on other variables changed association. Linear goodness of t was compared with exponential and quadratic transformations. Cohen's kappa was used for evaluating the visual IQ agreement and Spearman's rho for the correlation between the clinicians' visual IQ and weight, BMI, CV liv . All tests were two sided, with p<0.05 considered to be statistically signi cant.
[ 18 F]FDG PET/CT indication was in 80 to 82 percent of patients oncological (initial or follow-up exam in proven malignancy), versus diagnostic (benign versus malignant pathology) or miscellaneous (in ammatory or infectious pathology). On dPET/CT (with both imaging protocols) and aPET/CT, CV liv was associated with weight, BMI and fatty mass in univariable and multivariable linear regression analyses (p 0.0001 for dPET/CT and p 0.009 for aPET/CT). There was also a signi cant, more moderate association between sex and CV liv with higher CV liv in men on both camera's, only for dPET/CT 4MBq (0,11mCi) protocol 2 in multivariable weight models (p=0.03), and except for dPET/CT 3MBq (0,08mCi) protocol in multivariable BMI models (p 0.01).
Age, pathological exam, initial exam were not associated with CV liv .
For both readers there was a signi cant and similar negative relation of visual image quality score and BMI and weight (p 0001) on both camera's.  Visual analysis was not performed for 4MBq (0,11mCi)/kg protocol 2 on dPET/CT as semi-quantitative data were comparable in both protocols.

Discussion
In dPET/CT both imaging protocols with different acquisition time, injected activity and reconstruction parameters showed the same semi-quantitative relationship, extending its applicability.
To best of our knowledge this is the rst study of the impact of patient's habitus on PET image quality in digital PET/CT camera's, which is a very important step towards optimizing imaging protocols in these new camera's.
Digital PET/CT systems present an increased temporal resolution and stability in count rate [2] versus analogic PET/CT systems. However this doesn't compensate for the increase in noise and decrease of image quality with increasing patient's weight and habitus. In obese and overweight patients higher random counts, scatter activity besides attenuation and possibly pathologic hepatic heterogeneity may play a role in generating extra image noise.
A linear increase of the coe cient of variance, noise in the liver, was observed, with larger habitus and in particular weight for dPET/CT. However not signi cantly different, slightly higher tting of CV liv in function of weight versus BMI was obtained. Fatty mass was equally tted as weight but is not readily useable in clinical routine.
The linearity of the relation between patient's weight, BMI, fatty mass and CV liv is controversial in literature.
In multivariable analyses we also observed a moderate extra effect of sex with higher CV liv in men in some groups and regression models, without matching criteria for confounder.
Possibly increased incidence of metabolic syndrome in men with higher abdominal fat and/or attenuation and controversially higher heterogeneity in hepatic uptake in steatohepatitis [26], more frequent in men [27] could explain this difference. Obvious hepatic steatosis on CT (of PET/CT) was excluded, however no splenohepatic density comparison was used. We didn't measure abdominal waist or fat content.
On dPET/CT better IQ with lower mean CV liv values were obtained with 4MBq (0,11mCi) protocol versus 3MBq (0,08mCi) protocol, despite of a lower time activity product (4 versus 6). The association with weight was similar however. The PSF reconstruction was also altered from 2i10subsets for the 3MBq (0,08mCi) protocol to 4i4subsets for the 4MBq (0,11mCi) protocol, based on phantom studies and veri ed in clinical test patients.
There was a better IQ with lower CV liv in dPET/CT versus aPET/CT in 4MBq (0,11mCi)/kg protocols for patients<70kg and no signi cant difference of CV liv in intermediate weight category nor between correlation coe cients (Pearson, in weight and BMI models), despite of a BMI adaptive protocol, and with substantially longer mean scanning time and per step for all patients on aPET/CT. On dPET/CT also a smaller 2mm voxel size diameter was used versus 4mm on aPET/CT. Smaller voxel size increases image noise, by dividing detected number of photons, but improves small lesion detectability [28].
Similar negative correlation was found between visual [ 18 F]FDG PET image quality of both readers and patient's weight and BMI, on both camera's. The in uence of sex was however opposite on dPET/CT for global visual IQ (lesser quality in women) and not found in visual hepatic homogeneity scores.
CV liv was correlated with global visual IQ for both readers and even better with visual hepatic heterogeneity only.
Global visual IQ was besides liver homogeneity and image noise, image contrast and adequate visualization of low activity zones, also based on blinding artefacts around high activity zones (especially bladder) which were quite frequently seen on aPET/CT and not on dPET/CT. A more stable PET counting rate (with less dead time effects) in and around high activity areas is certainly an advantage of dPET/CT [2]. Unweighted Cohen's kappa for visual IQ between both readers was only moderate but identical on both camera's. Moreover one reader scored consistently a bit higher.
Study limitations are the retrospective and different study populations and protocols. Normal and slightly overweight patients were overrepresented, especially in the 4MBq (0,11mCi) protocol in dPET/CT.
Study evaluation parameters were based mainly on image noise and contrast in normal organs but don't take into account (small) lesion detectability.
Improvement of PET IQ in larger and higher weight patients and IQ harmonization are very important for exam accuracy, intrapatient follow-up (for instance in case of weight uctuations) as well as harmonization interpatient and intermachine in clinical routine, research and development of arti cial intelligence software.
The next step towards improvement of [ 18 F]FDG PET IQ in dPET/CT will be to evaluate an adjusted PET protocol.
Our aim will be to keep constant PET IQ with a CV liv of 0,122 0,008 corresponding to a and with a visual IQ score of at least 4 as second criterion, which has to be ful lled in at least 84% of our patients.
For this we can increase injected activity and/or acquisition time per bed position (in function of weight) and/or use technical solutions.
We will study two methods based rst on imaging denoising software and secondly an adaptive acquisition time/ bed position in function of weight categories. PSF reconstruction 4i4s is withheld. We chose not to include sex as a parameter given con icting semi-quantitative and visual data in this regard.
Imaging denoising software based on arti cial intelligence, like subtle PET TM by Subtle Medical is a very promising, FDA approved tool [29,30] which can even allow decrease in acquisition time and/or injected activity to keep a good and ideally constant IQ.The rst step would be to reevaluate all study patients retrospectively with this software. Second ideally a prospective study of both strategies will be done.
In the second strategy we prefer not to increase injected activity above 3MBq (0,08mCi)/kg mostly for radioprotection and practical reasons.
For the adaptive scan duration the conservative mean acquisition time per bed position can be calculated via the 3MBq (0,08mCi) protocol formula obtained by the univariable tted plot of CV liv versus weight (with the ancient reconstruction protocol) and the estimated effect of scan duration. We have however some incertainty about the level of in uence of the reconstruction versus the injected activity on IQ, rendering the formula rather conservative. The effect of acquisition time/bed position is based on phantom data veri ed in clinical test patients. Patients are categorized in (7)    Graph 2: Distribution of visual image quality scores in 3 weight categories on dPET/CT 3MBq (0,08mCi) protocol Only few patients on dPET/CT had a score 2 (N=3; 5%) or 5 (N=4; 7%) by at least one reader. resumeshort.xlsx