Coronavirus disease 2019 (COVID-19) in children: a systematic review of imaging findings

Background COVID-19 is a novel coronavirus infection that can cause a severe respiratory illness and has been declared a pandemic by the World Health Organization (WHO). Because children appear to be less severely affected than adults, their imaging appearances have not been extensively reported. Objective To systematically review available literature regarding imaging findings in paediatric cases of COVID-19. Materials and methods We searched four databases (Medline, Embase, Cochrane, Google Scholar) for articles describing imaging findings in children with COVID-19. We included all modalities, age <18 years, and foreign language articles, using descriptive statistics to identify patterns and locations of imaging findings, and their association with outcomes. Results Twenty-two articles were included, reporting chest imaging findings in 431 children, of whom 421 (97.7%) underwent CT. Criteria for imaging were lacking. At diagnosis, 143/421 (34.0%) had a normal CT. Abnormalities were more common in the lower lobes and were predominantly unilateral. The most common imaging pattern was ground-glass opacification (159/255, 62.4%). None of the studies described lymphadenopathy, while pleural effusions were rare (three cases). Improvement at follow-up CT imaging (3–15 days later) was seen in 29/100 (29%), remained normal in 25/100 (25%) and progressed in 9/100 (9%). Conclusion CT chest findings in children with COVID-19 are frequently normal or mild. Lower lobes are predominantly affected by patchy ground-glass opacification. Appearances at follow-up remain normal or improve in the majority of children. Chest CT imaging adds little to the further management of the patient and should be reserved for severe cases or for identifying alternative diagnoses. Electronic supplementary material The online version of this article (10.1007/s00247-020-04726-w) contains supplementary material, which is available to authorized users.


Introduction
A novel strain of coronavirus (referred to as 2019-nCoV or SARS-CoV-2), which causes the sometimes severe respiratory infection COVID-19, was first identified in Wuhan city, China, toward the end of 2019 [1]. By 12 March 2020, COVID-19 was declared a global pandemic by the World Health Organization (WHO), and at the time of writing it had spread to 187 countries, with almost 3.5 million confirmed cases, and had claimed more than 244,000 lives [2].
Epidemiological studies originating from China have shown that children are less likely to be clinically affected than older adults, with one study finding only 0.9% of those affected being younger than 15 years old [3]. A different Chinese study, which included 731 confirmed paediatric cases, found that the majority (84.1%) sustained either mild or moderate symptoms, with <3% described as being severely or critically affected [4]. Only one study reported a death in a child, a 10-month-old with multiorgan failure and intussusception [5]. In the United States, data published on 2 April 2020 [6] showed that only 1.7% (n=2,572) of people affected by COVID-19 were younger than 18 years, with 0.58-2% of children requiring intensive care admission. Given the low number of paediatric cases, keeping abreast of the latest information and assimilating the combined knowledge of radiographic findings in infected children is challenging. Whilst several systematic reviews of imaging findings in COVID-19 cases have been performed for adults [7,8], none has specifically focused on children.
The overall objective of this study was therefore to assimilate the available information on imaging features of COVID-19 disease in children. Particular points of interest include identifying typical imaging findings during diagnosis and follow-up stages of the infection, and whether any features might be used as prognostic markers to determine patient outcome. Where knowledge gaps exist, we intend to highlight these and suggest potential future avenues for research.

Materials and methods
Ethics approval was not required for this retrospective review of published data. We followed Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) guidelines for transparent reporting of systematic reviews. This study was registered in PROSPERO, an international prospective register of systematic reviews (Registration ID: CRD42020175945 [9]).

Literature review
A systematic literature search was performed of Medline (Ovid), Embase and the Cochrane Library databases for the latest articles published between 1 January 2015 and 17 March 2020 (5-year range), using database-specific Boolean search strategies with key terms and word variations relating to all three categories: (1) "coronavirus", "COVID-19", "SARS-CoV-2" or "2019-nCoV"; (2) "paediatrics", "children", "neonate", "infant" or "adolescent"; (3) "radiology", "imaging", "ultrasound", "CT", "MRI" or "radiography." Full search terms are shown in the supplementary material (Tables S1, S2 and S3). To include as many recent articles as possible, we also performed a grey literature search (i.e. literature not formally published in sources such as books or journal articles [10], such as government white paper articles and guidelines) using the same keywords on Google Scholar and for any WHO Global Library publications. Additional articles were retrieved by manual screening of the reference lists of included studies and relevant review articles/editorial pieces. The initial search was conducted on 17 March 2020. A repeat search was conducted on 23 March 2020, and again on 30 April 2020 for any further eligible manuscripts.

Eligibility criteria
Inclusion criteria encompassed all studies investigating and describing imaging findings of confirmed COVID-19 infection in children, using reverse transcriptase polymerase chain reaction (RT-PCR) testing. Studies were limited to human subjects, including foetuses (any gestation) and children (ages 0-18 years). No restrictions were placed on type of imaging modality, number of cases described or type of clinical setting. To widen our search and include as many cases as possible, we included case reports. No language restrictions were used given that many early articles on imaging findings have been published in Chinese. Where this was the case, we sought online translation services and advice from native-speaking colleagues to interpret reported findings.
Exclusion criteria included studies reported as editorials, opinion articles or multimedia files (online videos, podcasts). Suspected but unconfirmed cases of COVID-19 were not included. Studies relating to other coronavirus-related illnesses, such as Middle East respiratory syndrome (MERS) or severe acute respiratory syndrome (SARS) were excluded. We also excluded any articles reporting on a mixed adult and paediatric cohort where imaging results for the paediatric cohort could not be extracted.

Data extraction and quantitative data synthesis
All articles were independently searched by two reviewers (S.C.S. and S.T. with 7 years and 10 years of paediatric radiology experience, respectively). They examined abstracts of suitable studies and obtained full papers according to the eligibility criteria. Disagreements were resolved by consensus.
The same two reviewers (S.C.S., S.T.) independently extracted data from the full articles into a database (Excel; Microsoft, Redmond, WA), which included the following factors: study design, study setting/country, population demographics (e.g., gender, age, underlying comorbidities), sample size, patient outcomes (number of mortalities), imaging modality and imaging findings (pattern and location of involvement of disease) and results of any follow-up imaging.
Missing data were recorded as "not recorded" or "not stated". Authors of published studies were not contacted because of the tight time constraints involved in the completion of the systematic review during the unprecedented time of need for this information.

Methodological quality
The quality for each included study was assessed using the National Institutes of Health Quality Assessment Tool for Case Series Studies [11] by two reviewers (S.T. and J.L, with 17 years of paediatric radiology experience). Disagreements were resolved by consensus review. Any that could not be resolved by consensus was arbitrated by a third reviewer (S.C.S.).

Statistical analysis
We planned a meta-analysis to assess association of imaging findings with patient outcomes and demographic data; however, we omitted this because of a lack of sufficient data, with many cases being incompletely reported. Therefore, we used descriptive statistics to determine frequency and percentages of imaging appearances across different articles.

Included studies
During the initial literature search, after removing duplicates, we identified 146 unique records. After screening titles and abstracts, we excluded 100 studies and checked 46 full-text articles. Reasons for exclusions included insufficient description regarding imaging results (n=8), opinion pieces (n=7), adult population only (n=5), no confirmed (only suspected) COVID-19 cases (n=3), retracted article (n=2) and no full text available (n=1). After the second and third searches of the databases, we found eight more records that met our inclusion criteria and reviewed them.
Although it was not explicitly stated in the articles, we found that a small case series of 8 children [12] and a larger one of 115 children [13] appeared to describe a subset of a larger case series already published of 171 children with COVID-19 [5] (included in our review). All three articles originated from Wuhan city, China. We therefore excluded the two smaller studies from further qualitative or quantitative analysis. We have provided the reference and summary results from the smaller articles in our supplementary appendix (Tables S4 and S5; [5,) to alert readers to this overlap and show why we reached this conclusion.
Ninety-eight (98/431, 22.7%) children were asymptomatic upon admission to hospital, being assessed for having recent travel to a COVID-19 endemic area or close contact with an infected individual. Of those with symptoms, fever (120/333, 36.0%) and coughing (139/333, 41.7%) were the more common presenting complaints.

Initial imaging findings
Where CT imaging was used, 143/421 (34.0%) cases did not have any radiographic findings despite being COVID-19 positive. In the one study where only chest radiography was used, 6/10 (60%) of the imaged cases were normal [14]. In a case report from South Korea, the admission chest radiography was normal, although the CT demonstrated patchy nodular consolidation with ground-glass opacification [25]. Throughout all studies, there was significant heterogeneity in terminologies used with reference to radiographic findings, many of which included well-known terms (e.g., ground-glass opacifications) as well as non-standard descriptive terminologies (e.g., "thickened lung texture" [27]). In addition, missing information regarding imaging findings made summarising these difficult. Despite this, available findings are summated Tables 3 and 4 [5,. It should be acknowledged that the percentage of cases for the various descriptors is less important than their relative frequencies to one another, given the missing information.
From the reports where location of pathology was stated, this was identified in the upper lobe in 31/90 (34.4%), middle lobe in 15/90 (16.7%) and the lower lobe of the lung in 40/90 (44.4%) children. Diffuse/multifocal disease was described in 4 patients (4.4%).
Where a specific pattern was described, the most characteristic pattern on CT was ground-glass opacity, reported in 159/255 (62.3%) children. Patchy consolidations were seen in 14/255 (5.5%) patients. A so-called halo sign (of ground-glass opacification) around areas of consolidation was reported in 24/255 (9.4%) children across four studies [18,28,29,34]. Nonspecific terminologies of "lung infiltrates/shadows" were reported in 111/255 (43.5%) and interstitial lesions in 5/255 (1.9%) children. Not all articles had case-specific individual descriptions of chest radiography and CT findings, which were frequently described together.
Several findings were either not reported or only rarely reported. For example, none of the articles described significant mediastinal lymphadenopathy or cavitation on imaging, although only one study performed a contrast-enhanced CT. In only 3/255 (1.2%) children there were pleural effusions. Of these, one was a neonate also infected with respiratory O v e r a l lr a t i n g  (9) Were the results well-described?  [22]. The underlying conditions of the other cases were not reported [23].

Imaging findings and demographics
Given the small number of cases and the heterogeneous nature of case reporting, we could not to determine whether differences in imaging presentations varied significantly with patient age group, gender or presenting symptoms.
Regarding differences with adults, Ma et al. [23] found that children in their cohort (compared to a published adult cohort of 1,099 cases [3]), were more likely to demonstrate The findings correspond to readily available reported imaging findings within the relevant publications. Only COVID-19 confirmed cases by reverse transcriptase polymerase chain reaction (RT-PCR) are included. Because of the heterogeneous and occasionally incomplete reporting of these findings (e.g., some without pathology location, some using different terminologies), not all features are mutually exclusive, nor total to the combined number of patients; therefore, percentages are not provided for features that were only reported by a small number of publications (e.g., segmental involvement and subpleural distribution), and the denominator for lobe affected and laterality is derived for total number of studies where these findings were stated

Patient outcomes
At time of publication, 296/431 (68.7%) children had been discharged from the hospital, 72/431 (16.7%) were in a stable condition in a hospital, 3/431 (0.7%) had been admitted to intensive care units. One child admitted to intensive care reportedly later died at 4 weeks post hospital admission; this was a 10-month-old with multiorgan failure, septic shock and intussusception [5]. The outcome of the remainder of cases was unclear from the published reports (60/431, 13.9%).
One study performed subgroup analysis to determine the relationship between patient outcomes and CT imaging results [23]. The authors reviewed a subset of 23/50 (46.0%) children in their cohort who had been discharged, whose symptoms had resolved and who had negative RT-PCR testing on two separate occasions. The majority of these children 17/23 (73.9%) still demonstrated lung abnormalities, and in 2 cases these had progressed. A Cox regression analysis did not find any statistically significant association between differences in imaging findings during treatment and likelihood of discharge. This was supported in part by findings by Xia et al. [28], who stated that CT findings appeared to lag behind resolution of clinical symptoms and two sets of negative nucleic acid testing.

Discussion
Chest CT imaging findings in children with COVID-19 are frequently normal or mild, with lower lobes most commonly Descriptors refer to those stated within the relevant publications. Because of the heterogeneous, non-standard terminologies, we included descriptions of "shadows/infiltrates" together and interpret these to mean nonspecific opacities. In many articles, there was incomplete reporting of findings; therefore, not all features are mutually exclusive, nor total to the combined number of children within the study. The column titled "pattern described" is therefore included to demonstrate how many of the reported abnormal CT cases for which the study provided the abnormalities. The relative frequencies of findings are provided (with the "total pattern described" as denominator, rather than total abnormal CT), which is the more important indicator than the absolute numbers summated CT computed tomography, GGO ground-glass opacification affected, demonstrating patchy ground-glass opacification or, less frequently, areas of consolidation. Imaging appearances at follow-up remain normal or improve in the majority of children.
These findings raise important clinical implications for paediatricians and radiologists. Given the mild and sometimes absent findings on chest radiography and CT, it is unlikely that imaging will provide an increased confidence in the diagnosis for COVID-19, nor can it provide reassurance for the absence of infection if imaging is normal. This finding is supported by a recent study of 24 asymptomatic carriers of COVID-19, of whom 6 were children (5-15 years old) and all had normal chest CT findings [19]. Similar results have also been supported by work in adults [40,41]. Whilst it is well known that chest radiography can underplay the chest CT findings, the identification of mild to moderately severe imaging appearances in the majority of children, with little subsequent change in management, is unlikely to justify the CT imaging.
Conversely, even when CT findings are present, we have found that these can be heterogeneous and nonspecific, and in the adult literature it has been reported that differentiation between COVID-19 and other viral infections based on imaging findings is difficult [42]. Research on this topic in paediatric radiology is currently unavailable, although Qiu et al. [30] compared clinical findings between children with COVID-19, H1N1 influenza and severe acute respiratory syndrome (SARS) and found that children with COVID-19 had a much milder course of illness than with SARS, but were more likely to develop pneumonia than those with H1N1 influenza. Future studies that focus on attempting to differentiate imaging findings among these viral illnesses in children should be mindful to account for co-existing viral infections given that one publication in this study demonstrated COVID-19 in a child with co-infection of RSV [22].
In recent publications, it has been shown that some paediatric patients might be "super spreaders" (i.e. high viral shedding) [43] and more likely to demonstrate early symptoms. Whether they also display earlier CT changes, which could be used to screen for individuals prior to available RT-PCR testing, remains unknown. Therefore, given the lower severity of disease and additional radiation burden, CT imaging should not be routinely conducted for diagnosis, but rather reserved for those with severe or deteriorating symptoms, or in the search for an alternative diagnosis to aid management.
The fact that imaging appearances frequently resolve, improve or remain normal at follow-up imaging is reassuring because it suggests that long-term pulmonary damage is unlikely, although at present there is insufficient evidence to confirm this. It is also important to bear in mind that the persistence of pulmonary findings does not necessarily imply ongoing infection, given that one study found persisting CT findings in 17/23 (73.9%) children who had been treated, with resolution of their symptoms and two negative RT-PCR tests. Therefore, follow-up CT imaging would also be better guided by clinical symptoms rather than being performed as a matter of routine. These recommendations have recently been incorporated into newly published guidelines for imaging in  [44], which encourages a pragmatic approach based on patient symptom severity using chest radiography for initial workup and only for follow-up in moderate to severe cases, reserving an initial chest CT only where a clinical change in management is anticipated and for the post-recovery stages. Useful advice regarding general staff protection and departmental organisation for imaging of COVID-19 cases has also been recently provided by the European Societies of Radiology (ESR) and Thoracic Imaging (ESTI) [45]. It is worth noting that although our inclusion criteria were not set to review infected pregnant women, there was one case in this review of an infected newborn, diagnosed at 36 h of age from a COVID-19-positive mother [27]. The authors proposed the possibility for vertical transmission as a route of infection; however, several subsequent articles reviewing outcomes of infected pregnant women have now suggested this to be unlikely [39,[46][47][48]. A recently published rapid review of coronavirus in pregnancy [49] found that of the 32 infected women identified in the literature, there was 1 stillbirth (34 weeks of gestation) and 15/32 (47%) pre-term deliveries. In the 15/32 (47%) neonates who were tested for COVID-19, all were negative. The latest guidance from the Royal College of Obstetricians and Gynaecologists [50] has therefore recommended against routine separation of affected mothers and their babies and has not found any evidence to suggest intrauterine fetal infection or teratogenic effects from the novel coronavirus. Clinicians should thus remain alert to alternative, more common diagnoses in newborns presenting with respiratory symptoms (e.g., respiratory distress syndrome, aspiration, pneumonia from alternative organisms), even if the mother is COVID-19 positive [48].
Although we performed a comprehensive systematic review, there are still several gaps in our radiologic knowledge regarding COVID-19 in children. In this article we included manuscripts relating to radiographic or CT appearances of lung pathologies in children, although information on other modalities is lacking. There is sparse literature on the use of portable point-of-care ultrasound (POCUS) for adult COVID-19 patients in Italy [51] and in children [52,53]. In adults, POCUS has been reportedly used to triage more severe cases for urgent management by helping to identify areas of groundglass opacification (so-called B lines) as well as areas of necrotic lunga marker of the more advanced stages of infection [54]. Nevertheless, given that severe disease in children is less likely, the extent to which POCUS might be helpful is questionable, although it has been recommended as one of several potential options for lung assessment by a Chinese expert consensus review for neonatal management in COVID-19 [55]. Two adult publications have reported the use of [F-18]2-fluoro-2-deoxyglucose (FDG) positron emission tomography (PET)/CT in COVID-19 [56,57], and suggested that it could help monitor disease progression and treatment outcomes, particularly by detecting residual activity in mediastinal lymph nodes. This modality has not been reported in infected children, and adult studies have only included a small number of cases (4 patients or fewer). The added value is thus still undetermined [58] and should not be first studied in children, particularly given the increased radiation burden.
Our review also did not find many articles reporting imaging findings in immunocompromised children, so it is undetermined whether these features might differ from those without health conditions. It has been well documented that more severe infections are found in immunocompromised children with other strains of coronavirus [59] and that these infections can spread to other parts of the body, such as the brain, causing encephalitis [60]. Whether this also occurs with COVID-19 remains to be seen. It is interesting that the only child death reported in this review was also suffering from an intussusception. Whether this was triggered by the underlying viral infection is unclear, but it is worth noting that gastrointestinal complaints can be the first and more prominent symptom of COVID-19 in some people and has been reported in an adult series [61].
There are several limitations to this work. Given the widespread public health interest, several manuscripts on the topic of COVID-19 are emerging each week, many bypassing usual peer review processes. It is likely that some information might be missing but later included, or in certain cases articles retracted (as for two articles during our screening process). Nevertheless, where possible, we have tried to mitigate this by conducting our literature review thrice in order to provide the most up-to-date information from reliable sources. Whilst not all imaging findings in all cases were reported in the studies, we described all available findings to give a general overview of the imaging pathology. Future works on the study of COVID-19 imaging findings could be improved by the use of standardised detailed descriptors for imaging findings (i.e. stating both the pattern and localisation of findings), in line with RSNA guidance [44], with clearly stated indications for imaging where possible.
Second, because of the origin of the virus in China, some articles have been published in a language other than English, or in English by non-native speakers. This might have hampered our understanding and interpretation of the data, although we used online translation services where required. Whilst other systematic reviews on the topic of COVID-19 have excluded articles not written in English, we thought it was important to review as many foreign-language articles where possible to increase our collective knowledge base, particularly given the few reported paediatric cases.
Finally, the majority of articles have included children from China, in particular Wuhan city. It is unclear whether some of these paediatric cohorts overlap, although we did identify three papers where there was clear similarity in many of the patients described, and we avoided repetition of findings in summary results. It is also important to highlight that differences in indications for CT imaging in children might also exist (which could explain why Ma et al. [23] found slightly more abnormalities on CT in children than adults), but unfortunately these indications were not made clear in the publications.

Conclusion
Chest imaging findings in children with COVID-19 are frequently normal or mild, with unilateral changes that include patchy ground-glass opacification, commonly affecting the lower lobes. Imaging appearances at follow-up frequently remain normal or improve in the majority of children. Chest CT imaging adds little to the further management of the child and should be reserved for severe cases or for identifying alternative diagnoses. Further areas of research should include information on imaging and clinical characteristics in immunocompromised children with COVID-19, and information on long-term follow-up, particularly in the more severely affected children. We should therefore be prudent with the usage of CT, particularly at a time where resources are stretched, and only use it in the more vulnerable populations.