A case–control evaluation of pulmonary and extrapulmonary findings of incidental asymptomatic COVID-19 infection on FDG PET-CT

Objectives: To describe the findings of incidental asymptomatic COVID-19 infection on FDG PET-CT using a case–control design. Methods: Incidental pulmonary findings suspicious of asymptomatic COVID-19 infection on FDG PET-CT were classified as a confirmed (positive RT-PCR test) or suspected case (no/negative RT-PCR test). Control cases were identified using a 4:1 control:case ratio. Pulmonary findings were re-categorised by two reporters using the BSTI classification. SUV metrics in ground glass opacification (GGO)/consolidation (where present), background lung, intrathoracic nodes, liver, spleen and bone marrow were measured. Results: 7/9 confirmed and 11/15 suspected cases (COVID-19 group) were re-categorised as BSTI 1 (classic/probable COVID-19) or BSTI 2 (indeterminate COVID-19); 0/96 control cases were categorised as BSTI 1. Agreement between two reporters using the BSTI classification was almost perfect (weighted κ = 0.94). SUVmax GGO/consolidation (5.1 vs 2.2; p < 0.0001) and target-to-background ratio, normalised to liver SUVmean (2.4 vs 1.0; p < 0.0001) were higher in the BSTI 1 & 2 group vs BSTI 3 (non-COVID-19) cases. SUVmax GGO/consolidation discriminated between the BSTI 1 & 2 group vs BSTI 3 (non-COVID-19) cases with high accuracy (AUC = 0.93). SUV metrics were higher (p < 0.05) in the COVID-19 group vs control cases in the lungs, intrathoracic nodes and spleen. Conclusion: Asymptomatic COVID-19 infection on FDG PET-CT is characterised by bilateral areas of FDG avid (intensity > x2 liver SUVmean) GGO/consolidation and can be identified with high interobserver agreement using the BSTI classification. There is generalised background inflammation within the lungs, intrathoracic nodes and spleen. Advances in knowledge: Incidental asymptomatic COVID-19 infection on FDG PET-CT, characterised by bilateral areas of ground glass opacification and consolidation, can be identified with high reproducibility using the BSTI classification. The intensity of associated FDG uptake (>x2 liver SUVmean) provides high discriminative ability in differentiating such cases from pulmonary findings in a non-COVID-19 pattern. Asymptomatic COVID-19 infection causes a generalised background inflammation within the mid-lower zones of the lungs, hilar and central mediastinal nodal stations, and spleen on FDG PET-CT.


INTRODUCTION
The coronavirus disease 2019 (COVID-19) pandemic has created the biggest global health crisis in generations. The spread of infection has been difficult to control due to asymptomatic infection, estimated to account for over 50% of all transmissions. 1 Nasopharyngeal swab reverse transcriptase polymerase chain reaction (RT-PCR) is considered the gold standard for diagnosing COVID-19 infection. 2,3 The sensitivity of RT-PCR in symptomatic patients ranges between 82 and 97% 4 but detection rates are lower in asymptomatic individuals. 5,6 During the early stages of the pandemic in 2020, numerous studies described computed tomography (CT) features characteristic of COVID-19 infection with some suggesting sufficient diagnostic accuracy of CT in the absence of RT-PCR testing 7,8 ; significant selection bias and several confounding factors have since undermined such conclusions. 9 International guidelines and a recent umbrella review recommend CT as a problem-solving tool to identify complications of COVID-19 infection or when an alternative diagnosis is suspected in symptomatic individuals. 10, 11 The British Society of Thoracic Imaging (BSTI) published discriminated between the BSTI 1 & 2 group vs BSTI 3 (non-COVID-19) cases with high accuracy (AUC = 0.93). SUV metrics were higher (p < 0.05) in the COVID-19 group vs control cases in the lungs, intrathoracic nodes and spleen. Conclusion: Asymptomatic COVID-19 infection on FDG PET-CT is characterised by bilateral areas of FDG avid (intensity > x2 liver SUV mean ) GGO/consolidation and can be identified with high interobserver agreement using the BSTI classification. There is generalised background inflammation within the lungs, intrathoracic nodes and spleen. Advances in knowledge: Incidental asymptomatic COVID-19 infection on FDG PET-CT, characterised by bilateral areas of ground glass opacification and consolidation, can be identified with high reproducibility using the BSTI classification. The intensity of associated FDG uptake (>x2 liver SUV mean ) provides high discriminative ability in differentiating such cases from pulmonary findings in a non-COVID-19 pattern. Asymptomatic COVID-19 infection causes a generalised background inflammation within the mid-lower zones of the lungs, hilar and central mediastinal nodal stations, and spleen on FDG PET-CT. (Continued) criteria for the diagnosis of COVID-19 infection on CT, based on the presence and distribution of ground glass opacification (GGO), consolidation and varied patterns of organising pneumonia (OP). 12 Asymptomatic individuals can have normal lungs on CT or alternatively demonstrate radiological features compatible with COVID-19 infection. [13][14][15] Several case reports/series of asymptomatic COVID-19 infection on 2-deoxy-2-[ 18 F]fluoro-D-glucose (FDG) positron emission tomography (PET)-CT report metabolically active findings mainly confined to the lungs and mediastinal lymph nodes [16][17][18][19][20][21][22][23][24][25][26][27] ; most studies have been purely descriptive, however. A few studies reporting increased FDG uptake in extrathoracic nodes, spleen, and bone marrow, suggest that FDG PET-CT can demonstrate the immune response to viral infections. 25  PET-CT in patients with suspected asymptomatic COVID-19 infection scanned during the 'first wave' of UK pandemic, are significantly different to those in a control group scanned prior to the pandemic, matched for age, gender and scan indication. We will also determine the ability of FDG uptake in conjunction with pulmonary findings categorised using the BSTI classification to discriminate between COVID-19 and non-COVID-19 infection, whilst assessing the interobserver agreement between two reporters using the BSTI classification.

Case selection
Institutional review board approval was obtained for this retrospective non-interventional observational case-control study. Inclusion criteria were: • FDG PET-CT examination performed between 23/03/2020 and 29/05/2020 during the 'first wave' of the UK pandemic. • Absence of new continuous cough or high temperature, i.e., asymptomatic.    • Expedited (via email notification) FDG PET-CT report due to incidental pulmonary findings suspicious for asymptomatic COVID-19 infection.
Referring clinicians either opted for confirmation of COVID-19 infection via RT-PCR hospital testing or recommendation for a period of self-isolation for 14 days as per UK government guidance due to a lack of community testing at the time. 30 Patients with a positive RT-PCR test within 28 days of scanning were classified as a confirmed case. Patients with pulmonary findings suspicious of COVID-19 infection on FDG PET-CT but no or negative RT-PCR test within 28 days of scanning were classified as a suspected case. Information on RT-PCR testing and clinical follow-up was obtained from institutional electronic databases. Consecutive control cases matched for age, gender and scan indication without exclusion criteria were identified from spring 2019 ± 3 months, using a 4:1 control: case ratio.
FDG PET-CT imaging review FDG PET-CT examinations were performed using methodology aligned to EANM guidance and described previously. 31 All examinations were anonymised (including date of examination) and analysed using Hybrid Viewer (Hermes Medical Solutions, Sweden). Independent blinded review of pulmonary findings was undertaken 6 months after the 'first wave' by board certified radiologists (S.B) and consultant radiologist (M.S) with 1 and 10 years of PET-CT reporting experience, respectively, and each with 12 years of diagnostic CT (including thoracic CT) reporting experience. Pulmonary findings were categorised using the BSTI classification 12 ; classic/probable COVID-19 (BSTI 1), indeterminate COVID-19 (BSTI 2), non-COVID-19 (BSTI 3), and normal (BSTI 4). The normal category (BSTI 4) included findings considered within the spectrum of normality for PET-CT, e.g., gravity-dependent GGO and basal linear atelectasis. Following independent review, examinations with disagreement in BSTI classification had consensus reads. For examinations with clinically significant pulmonary parenchymal findings, i.e., BSTI 1-3, the highest maximum standardised uptake value (SUV max ) in an area of GGO/consolidation was documented, enabling targetto-background ratio (TBR) calculation, normalised to the mean standardised uptake value (SUV mean ) in the liver.
SUV metrics were derived from normal lung and from extrapulmonary sites (intrathoracic nodes, liver, spleen and bone marrow) by a consultant nuclear medicine physician (B.M.F) with 15 years of PET-CT reporting experience. Freehand regions of interest (ROIs) following the contours of the lungs but excluding subpleural regions and avoiding major vessels or parenchymal abnormalities were drawn in the upper (level of suprasternal notch), mid (1 cm below the carina) and lower zones (2.5 cm above the right hepatic dome) of both lungs to calculate SUV mean of background lung. ROIs were drawn around the major intrathoracic nodal stations 32 to calculate nodal SUV max ; SUV max was only measured if lymph nodes were visible on CT. Spherical volumes of interest (VOIs) were placed in the right lobe of the liver (6 cm diameter), spleen (3 cm diameter) and L4 vertebral body (2 cm diameter) as a representation of marrow uptake, to calculate SUV mean in these VOIs.

Statistical analysis
Interobserver agreement using the BSTI classification was assessed using the weighted κ method. 33 Non-parametric tests were used to assess for group-wise (Kruskal-Wallis) and pairwise (Mann Whitney U) differences. The Benjamini-Hochberg method to estimate the false discovery rate (FDR) was used to correct for multiple comparisons; an FDR < 0.05 was considered significant. Receiver operating characteristic (ROC) curves were generated with an area under the curve (AUC) calculated for each ROC 34 with the best thresholds for group discrimination defined using Youden's method. 35 All analyses were performed in R v. 4.0.0 with the base and stats packages while ROC analyses  were performed using the pROC package. More detailed information can be found in the Supplementary Material 1   (range 285-392 MBq), mean uptake time: 63 ± 4.5 min (range 56-78 min)).
Pairwise comparisons across the BSTI classification revealed no differences in SUV max GGO/consolidation (p = 0.088) or TBR (p = 0.064) between BSTI 1 and 2 cases. SUV max GGO/consolidation  Figures 8 and 9). SUV max GGO/consolidation ROC analysis ROC curves indicated excellent discrimination using SUV max GGO/consolidation with an AUC of 0.93 (0.84-1.00) for differentiating between the BSTI 1 & 2 group and BSTI 3 cases and 0.87 (0.75-0.99) between the COVID-19 group and control cases (Figure 10). Using a SUV max 3.15 cut-off, discrimination between the BSTI 1 & 2 group and BSTI 3 cases was achievable with a sensitivity of 0.90 and specificity of 0.88, whilst a SUV max 3.45 cut-off enabled discrimination between the COVID-19 group and control cases with a sensitivity of 0.75 and specificity of 0.95.

SUV metrics in pulmonary and extrapulmonary sites
There were significantly higher SUV metrics (p < 0.05) in the COVID-19 group vs control cases in 9/15 regions; 3/6 pulmonary regions (right mid zone, right lower zone, left lower zone), 5/6 nodal regions (bilateral hilar, bilateral paratracheal and subcarinal mediastinal nodal stations) and in the spleen. There was no significant difference in SUV mean in the liver or bone marrow (Supplementary Table 2, Supplementary Figure 1).

DISCUSSION
3.3% (24/732) of FDG PET-CT examinations performed during spring 2020 had incidental pulmonary findings suspicious of asymptomatic COVID-19 infection, which is within the quoted incidence range (2.1-16.2%) from a systematic review of 11 studies. 36 Our incidence is lower than reported in a study from a similar sized London institution (9.4%), 19 but this may be related to potential false-positive observations secondary to unilateral rather bilateral pulmonary findings, i.e., indeterminate for COVID-19 (BSTI 2) coupled with most of their cases with thoracic findings on PET-CT, either negative (4/15) or without RT-PCR confirmation (10/15).
Blinded consensus pulmonary analysis performed 6 months later, with a greater experience of reporting COVID-19 infection on FDG PET-CT, categorised 18/24 of the confirmed and suspected cases as either BSTI 1 (classic/probable COVID-19) or BSTI 2 (indeterminate for COVID-19), whilst none of the control cases from 2019 were categorised as BSTI 1. This confirms that the pattern of FDG avid pulmonary parenchymal changes observed in spring 2020 (Figures 1-3) was a novel phenomenon not experienced before, and also demonstrates the high specificity (97.9%) achieved through using the BSTI classification.
Several studies report an increased incidence of pulmonary findings suspicious for COVID-19 infection during the 'first wave' compared to control cases [20][21][22]24 similar to ours, except that patterns compatible with COVID-19 interstitial pneumonia were observed in their control cohorts; this is likely due to the presence of COVID-19 mimics on FDG PET-CT, e.g., influenza pneumonia or OP related to connective tissue disease or drug toxicity. However, most studies did not use a standardised CT grading system for categorising pulmonary changes, likely reducing specificity. Maurea et al 20 using the COVID-19 Reporting and Data System (CORADS) 37 reported 14/335 (4%) control cases with 'abnormal PET-CT findings suspicious for  COVID-19 infection' . However, 9/14 (64%) were classified as CO-RADS 2 (CT abnormalities consistent infection other than  or CO-RADS 3 (uncertain CT findings for  suggesting an overestimation of PET-CT findings suspicious for COVID-19 infection in their control population. Our study, the first to formally assess agreement between two reporters using the BSTI classification on FDG PET-CT, demonstrated almost perfect agreement (weighted κ = 0.94), even when BSTI 4 cases were excluded from the analysis (weighted κ = 0.83). Inui et al, 38 compared different CT grading systems for COVID-19 infection, and showed that all had reasonable diagnostic performance (0.80-0.84), albeit with lower interobserver agreement (Cohen κ = 0.61-0.63) than ours, but that CO-RADS and BSTI outperformed the other two classifications. Our higher interobserver agreement may be augmented by amalgamation of the 'classic' and 'probable' COVID-19 categories to represent BSTI 1 as per the published COVID-19 CT reporting proforma 12 rather than interpretate them as two separate categories. The inadequacies of the low-dose non-breath hold CT component of a PET-CT examination, requiring a more pragmatic approach to assessing the lungs, i.e., forgoing subtleties, also likely contributed to more consistent and reproducible observations. FDG uptake in areas of GGO/consolidation was significantly higher in the COVID-19 group vs control cases (SUV max 4.7 vs 2.1) and BSTI 1 & 2 group vs BSTI 3 cases (SUV max 5.1 vs 2.2); similar values have been reported in an early systematic review of incidental COVID-19 infection on FDG PET-CT (mean SUV max 4.9), 39 and Italian multicentre study (mean SUV max 4.1). 24 TBR analysis demonstrated that the intensity of FDG uptake in GGO/ consolidation was >x2 liver SUV mean in the COVID-19 and BSTI 1 & 2 groups, and lower and comparable to liver SUV mean for control and BSTI 3 cases. These findings confirm that a distinctive feature of COVID-19 infection is the association of high FDG uptake with areas of GGO/consolidation, related to multinucleated giant cells and focal clusters of lymphomonocytic infiltration in the context of diffuse alveolar damage, demonstrable even in early COVID-19 infection. 40,41 The discriminative ability of SUV max in areas of GGO/consolidation to differentiate between the BSTI 1 & 2 group and BSTI 3 cases was high; using a SUV max 3.15 cut-off, discrimination between BSTI 1 & 2 group and BSTI 3 cases was achievable with high sensitivity and specificity. From a clinical viewpoint, pulmonary findings compatible with classic/probable COVID-19 (BSTI 1), e.g., bibasal peripheral GGO/consolidation with 'reversehalo' or perilobular opacity, i.e., OP pattern, or indeterminate COVID-19 (BSTI 2), e.g., unilateral, non-peripheral GGO/ consolidation, are likely to have higher levels of associated FDG uptake (SUV max >3.15) in comparison with GGO/consolidation in a non-COVID-19 (BSTI 3) pattern.
Our study reaffirms that the low-dose non-breath hold CT component of the study can enable diagnosis despite not being of 'diagnostic quality' and is not solely for the purposes of attenuation correction and localisation. The absence of a full inspiratory effort and breathing artefact during scanning can limit accuracy, however. Difficulty in detection/characterisation of smaller lesions particularly towards the lung bases limits sensitivity, whilst an increased incidence of dependent GGO alongside areas of basal atelectasis, can be potentially misinterpreted as significant pathology, reducing specificity. Unrelated pulmonary pathologies, e.g., other viral pneumonias, OP secondary to connective tissue disease or drug toxicity or active pulmonary fibrosis can have similar appearances to COVID-19 infection, and will reduce specificity, although in our study this was only encountered in 2/96 control cases (Figure 4).
We found significantly higher SUV metrics in the COVID-19 group vs control cases in the mid-lower lung zones, both hilar and central mediastinal nodal stations, and spleen, suggesting the presence of generalised background inflammation. Lower zone predominant background pulmonary inflammation correlates with the tendency of COVID-19 infection to present with bilateral abnormalities affecting both lower lobes. 42,43 FDG avid intrathoracic and supraclavicular nodes with COVID-19 infection have been reported in several studies with varied frequency, with or without CT enlargement, 20,[22][23][24][25] whilst only one study has reported increased splenic uptake (5/13 patients) and increased bone marrow uptake 25 ; extrapulmonary abnormalities involving the salivary glands 19 and gastro-intestinal tract 18 were not routinely assessed for during our study. The presence of generalised systemic inflammation has been confirmed in small cohorts of patients recovering from COVID-19 infection (lungs, mediastinal nodes, spleen, liver, large vessels), 44,45 as well as in patients with post-COVID syndrome 46 in conjunction with findings of brain hypometabolism. 47,48 The major limitation to our study, common to many, is the absence of RT-PCR testing for all FDG PET-CT examinations during spring 2020, due to a lack of testing capacity. 30 Patients with COVID-19 infection but without pulmonary findings suspicious of infection will have been missed using our methodology, which will undoubtedly affect the sensitivity and specificity estimate of the BSTI classification on FDG PET-CT. In addition, this limitation also brings into question the classification of suspected cases which had either no (12/15) or a negative (3/15) RT-PCR test despite the presence of pulmonary findings suspicious of COVID-19 infection on FDG PET-CT (Supplementary Table 1). However, SUV max GGO/consolidation and TBR analysis demonstrated that confirmed and suspected cases were similar to each other but were individually as well as in combination (COVID-19 group), distinct from control cases, supporting our methodology of combining these cases for analysis (Tables 3 and 4, Figures 8  and 9). In addition, the sensitivity of the gold standard nasopharyngeal RT-PCR in symptomatic individuals is not 100% 4 and is lower in asymptomatic individuals, i.e., higher false-negative rates. 5,6 Sensitivity can be improved through repeat RT-PCR testing 49 or with bronchoalveolar lavage; a study of 46 patients reported 18 patients (39%) had a positive bronchoalveolar lavage RT-PCR despite two preceding negative nasopharyngeal RT-PCR tests with importantly 13 of these 18 patients (72%) with two preceding negative nasopharyngeal RT-PCR tests having CT findings compatible with COVID-19 infection. 50 This confirms the imperfection of single/multiple nasopharyngeal RT-PCR tests and that pulmonary changes compatible with COVID-19 infection in the context of a negative RT-PCR test(s) cannot be readily dismissed.

CONCLUSION
Asymptomatic COVID-19 infection on FDG PET-CT is characterised by bilateral areas of GGO/consolidation that are associated with increased FDG uptake (>x2 liver SUV mean ) and which can be identified with high reproducibility using the BSTI classification. These changes occur on the background of generalised inflammation within the mid-lower zones of the lungs, hilar and central mediastinal nodal stations, and spleen. This analysis will enable better preparedness for identification of asymptomatic COVID-19 infection on FDG PET-CT, prompting early confirmation RT-PCR testing, and minimising the risk of undetected infection to both the individual and society as a whole.
AVAILABILITY OF DATA AND MATERIALS: Anonymised data that supports these findings are available from the corresponding author upon reasonable request.