Molecular Imaging Findings on Acute and Long-Term Effects of COVID-19 on the Brain: A Systematic Review

Encephalitis

There were 3 independent case reports including ¹⁸F-FDG PET in patients with autoimmune encephalitis in the acute to subacute phase of COVID-19 (22–24). Given the preliminary nature of case reports, we are summarizing only the most noteworthy findings and refer the interested reader to Supplemental Table 1 for more detail. PET showed diffuse cortical hypometabolism (n = 1) and increased metabolism in the basal ganglia (n = 3), mesiotemporal structures (n = 1), and cerebellum (n = 1), possibly rated to cerebellum- and basal ganglia–specific antineuronal antibodies in 2 of these patients (22,24). Interestingly, immunomodulating treatment led to improvement in all patients, with 6-mo follow-up ¹⁸F-FDG PET having normal findings in 1 patient (24). The authors postulated para- or postinfectious encephalitis without evidence of direct virus invasion, which fits into other relative rare cases in the literature (13).

Parkinsonism and Other Neurodegenerative Diseases

Three case reports described molecular imaging findings in 4 patients in whom the first manifestations of parkinsonism occurred only weeks after SARS-CoV-2 infection (in part with other neurologic symptoms such as myoclonus, fluctuating consciousness, ocular abnormalities, and cognitive impairment; Supplemental Table 1). Again, these findings have to be regarded as preliminary although raising the important question of whether and how a COVID-19–related CNS pathology may unmask or worsen neurodegeneration (21): imaging of dopamine transporter availability with ¹²³I-FPCIT SPECT (25,26) and dopamine synthesis and storage with ¹⁸F-FDOPA PET (27) in 3 of these patients confirmed a nigrostriatal dopaminergic deficit. In 1 patient, normal cardiac innervation as assessed by ¹²³I-MIBG scintigraphy and subsequent improvement without specific treatment argued against incipient Parkinson disease. The authors speculated that SARS-CoV-2 virus may have gained access to the CNS and affected the midbrain (25). Two patients showed clinical and ¹⁸F-FDG PET findings compatible with postinfectious immune-mediated encephalitis that bore similarities to the encephalitis lethargica that was seen after the influenza epidemic in 1918 (26). However, unlike encephalitis cases described in the paragraph above, these patients did not improve (neither spontaneously, nor with immunomodulatory treatment), and it is unclear whether parkinsonism might have been caused by subsequent development or unmasking and hastening of preclinical neurodegeneration (26). Similarly, a patient reported by Cohen et al. (27) was diagnosed with probable Parkinson disease according to current diagnostic criteria, possibly facilitated by genetic susceptibility or virus-induced inflammation. To the best of our knowledge, no imaging follow-up has yet been published, which would help to unravel the underlying mechanism. Finally, the possibility that SARS-CoV-2 infection may precipitate and accelerate neurodegenerative diseases was also raised by Young et al. (28), who presented the case of a previously healthy man who developed rapidly progressive and fatal Creutzfeldt–Jakob disease 2 wk after being diagnosed with COVID-19. In line with Creutzfeldt–Jakob disease, ¹⁸F-FDG PET showed left hemispheric hypometabolism. The authors referred to animal and human studies showing that proinflammatory cytokines may promote neuroinflammation and progression of various neurodegenerative forms of dementia and parkinsonism (28).

Focal Symptoms or Lesions

Olfactory dysfunction is the most frequent focal neurologic sign in COVID-19. Several case reports (Supplemental Table 1) and 2 systematic studies (Supplemental Table 2) addressed this symptom. Niesen et al. (29) prospectively examined 12 patients with ¹⁸F-FDG PET/MRI about 2 wk after a sudden loss of smell in COVID-19. Single-subject and group analyses of ¹⁸F-FDG PET data showed no significant regional findings when using a conservative statistical threshold (i.e., P < 0.05, familywise error [FWE]–corrected). When using a liberal threshold (P < 0.001), individual analyses showed a heterogeneous pattern of metabolic decreases (n = 3), increases (n = 1), or both (n = 8) in various regions (e.g., olfactory regions, primary and higher-order cortices, and cerebellum), whereas group analyses resulted in clusters of hypometabolism in medial and dorsal frontal areas and hypermetabolism in orbitofrontal and parietal cortex and thalamus. Given the heterogeneity of findings, the authors concluded that the main pathophysiologic hypotheses of COVID-19–related hyposmia (i.e., olfactory cleft obliteration and neuroinvasion of SARS-CoV-2) do not explain dysosmia in all patients and that the PET findings probably reflect deafferentation and functional reorganization (29). In contrast, earlier case reports found orbitofrontal hypometabolism in patients with anosmia (30,31), which was interpreted as a result of direct neuroinvasion of SARS-CoV-2 via the olfactory bulb. As preliminary support for a possible extension of impairment to other brain structures, the latter groups also reported hypometabolism of medial temporal structures in a patient with COVID-19–related parosmia (32) and a patient without olfactory dysfunction (31) (in addition to other regions; Supplemental Table 1). A recent systematic study also found hints of an impairment of mesiotemporal structures in COVID-19–related olfactory dysfunction: Donegani et al. (33) prospectively recruited 14 patients with isolated hyposmia during the recovery phase from COVID-19 (4–12 wk after the first positive PCR result). Compared with control subjects and applying corrections for covariates (age, sex, scanner type) and multiple comparisons (i.e., P < 0.05, FWE-corrected), patients with isolated hyposmia were characterized by clusters of hypometabolism bilaterally in the parahippocampal and fusiform gyri and in the left insula, possibly reflecting cortical deafferentation. Beyond its implication in hyposmia, the involvement of limbic regions in COVID-19 may imply a risk of developing long-term neurologic (possibly cognitive) impairment, as discussed by the authors (33).

Other focal signs preliminarily investigated by ¹⁸F-FDG PET include a case of facial palsy with putative hypometabolism of the respective facial nerve (34) and a patient with frequent focal seizures possibly due to subacute encephalitis after SARS-CoV-2 infection with a normal ¹⁸F-FDG PET result (35) (Supplemental Table 1). Another group reported hypermetabolism of the inferior colliculi as a novel finding in patients with COVID-19 (36,37) that was associated with more frequent seizures and higher blood leukocytes at admission. However, the cause of this finding (e.g., inflammatory reaction (37), hyperactivation (36), and artifact) and its clinical relevance needs to be defined (Supplemental Tables 1 and 2).

Encephalopathy

All patients reported in this “Encephalopathy” section underwent inpatient and possibly intensive care unit treatment because of the overall high severity of COVID-19 symptoms. Furthermore, symptoms compatible with encephalopathy occurred with or briefly after the onset of general COVID-19 symptoms. Thus, the following study populations comprise a fairly homogeneous group. This homogeneity probably constitutes the basis for the observation that available ¹⁸F-FDG PET studies in COVID-19–associated encephalopathy yielded very consistent results (Table 1; Supplemental Table 1):

An initial case series by Delorme et al. ((38); n = 4) and subsequent systematic prospective studies by Kas et al. ((39); n = 7) and Hosp et al. ((18); n = 15/29 with PET, exhibiting symptoms compatible with encephalopathy) unanimously showed that the acute to subacute phase (±1 mo after infection) of COVID-19–associated encephalopathy is characterized by cognitive impairment (e.g., psychomotor agitation or slowing, executive functions, attention, visuoconstruction, and memory) and occasional other neurologic signs (e.g., hemiparesis, ataxia, apraxia, aphasia, myoclonus, and seizures). This clinical presentation is accompanied by a pronounced hypometabolism of frontoparietal-dominant neocortical areas (visual reads and conventional statistical parametric mapping [SPM] analyses, P < 0.05, FWE-corrected, or P < 0.01, false-discovery rate–corrected). Similar results were gained from a voxelwise, principal-components analysis–based comparison to age-matched controls, which yielded an extensive pattern of negative voxel weights (i.e., reduced metabolism) in frontoparietal-dominant neocortical areas and the caudate nucleus (18). Although an apparent hypermetabolism of striatum, pons, and cerebellar structures was found on individual visual reads and in statistically liberal (P < 0.05, uncorrected) group analyses (18,38,39), more thorough analyses suggest that these regions most likely show a preserved, actually not elevated, metabolism (18,40). Figure 1 shows the results of principal-components analysis and SPM group comparisons of patients with COVID-19–related encephalopathy compared with a group of healthy controls. Moreover, the individual expression score of the COVID-19–related spatial covariance pattern exhibited a highly significant, negative correlation with individual results from cognitive testing using the Montreal Cognitive Assessment (r² = 0.62, P < 0.001 (18)).

• Download figure
• Open in new tab
• Download powerpoint

FIGURE 1.

¹⁸F-FDG PET in COVID-19–related CNS disorders: principal-components analysis of spatial covariance pattern (first row) and SPM analysis of metabolic group differences (second to fifth rows) in patients with COVID-19–related encephalopathy (n = 15: first and second rows, Hosp et al. (18); of them n = 8 at 6-mo follow-up examination in third row, Blazhenets et al. (41)), patients with post–COVID-19 syndrome (n = 14; fourth row), and patients with post–COVID-19 syndrome and hyposmia (n = 9: fifth row, Dressing et al. (46)). Columns from left to right are lateral (left/right), superior, and mesial (right/left) views of cerebrum and lateral (left/right) views of brain stem and cerebellum (overlay created with Surf Ice 2 software; https://www.nitrc.org/projects/surfice/). Each analysis was performed in comparison to healthy controls (n = 13; 7 men and 6 women; mean age, 68 ± 7 y; PET performed under strictly comparable conditions), including age as covariate. SPM analyses entail count rate normalization to white matter (Hosp et al. (18); virtually identical results were gained with scaling to pons, whereas scaling to total brain parenchyma resulted in apparent [artificial] hypermetabolism in subcortical structures in COVID-19–related encephalopathy (40)). SPM results (t maps) were thresholded liberally for comprehensive display of findings (t = ±2, corresponding to P ≈ 0.05, uncorrected). Only extensive neocortical hypometabolism in COVID-19–related encephalopathy survives correction for multiple comparisons (voxel threshold, P < 0.05, false discovery rate–corrected).

Sequential follow-up data 1 mo later (39) and at 5–6 mo after infection (39,41) demonstrated a steady, though not necessarily complete, improvement of the clinical condition and ¹⁸F-FDG PET findings in most patients: in the study by Kas et al. (39), all but 1 patient showed a normal neurologic examination and recovered autonomy of daily living on follow-up. Still, all had abnormal cognitive evaluations with at least attentional or executive deficits, and 4 presented with psychiatric impairments. In parallel, cerebral metabolism improved on a group level, with mild residual hypometabolism of the left and right rectal gyri, the right insula, and the caudate nucleus and cerebellum. On a subject level, all but 1 patient, who got worse, showed a moderate to almost complete improvement (39).

Blazhenets et al. (41) provided the follow-up data of 8 patients from the COVID-19 register described by Hosp et al. (Table 1) (18): in parallel to a significant improvement in Montreal Cognitive Assessment performance (from 19 ± 5 to 23 ± 4, P = 0.03; still below normality [<26/30] in n = 5/8), follow-up ¹⁸F-FDG PET examinations showed a significant recovery (P < 0.01, false-discovery rate–corrected) of initial neocortical hypometabolism, with small areas of reduced metabolism being still detectable at a liberal statistical threshold (P < 0.005) at 6 mo. Likewise, the pattern expression score of the COVID-19–related spatial covariance pattern also significantly decreased over time (P < 0.005), albeit being still higher than in controls at a trend level (P = 0.06). Again, the pattern expression score correlated inversely with cognitive performance (repeated-measure r² = 0.39, P < 0.01 (41)).

The authors of aforementioned studies agreed that COVID-19–related encephalopathy and the accompanying PET findings are unlikely to be caused by encephalitis due to neuroinvasion of SARS-CoV-2. The findings are more likely explained by a parainfectious systemic immune mechanism (e.g., cytokine release (18,38,39,41)), which according to an autopsy case in the study by Hosp et al. (18) manifests as pronounced microgliosis in the white matter, whereas the gray matter was mostly spared. Furthermore, it was suggested that an underlying neurodegenerative disease may be a predisposing factor (38), particularly in those who do not recover.

Post–COVID-19 Syndrome

Post–COVID-19 syndrome is defined by persisting symptoms at 3 mo after symptom onset. Thus, it may include a highly variable group of patients in terms of initial disease severity, temporal course, and complaints. Patients still recovering from earlier manifestations (e.g., encephalopathy, usually treated as inpatients) may fall into this category, as will those who experience subjective complaints of brain fog only weeks after infection. Up to now, 1 preliminary case report (Supplemental Table 1) (42) and 4 systematic studies have been published (Table 2).

The team of Guedj et al. reported 2 studies on adults ((43); n = 35) and children ((44); n = 7) with persistent subjective complaints (mostly memory or cognitive complaints, fatigue, and insomnia) after apparent recovery from COVID-19. The mean or median delay between COVID-19 onset and PET was 3 and 5 mo in the adult and pediatric population, respectively. However, the time spans (26–155 d and 1–8 mo, respectively) indicate that both studies included subjects with ongoing COVID-19 according to the definitions (5) (they would also not fulfill the novel definition of post–COVID-19 condition by the WHO (20)). Both studies provided similar results (P < 0.05, FWE-corrected, in adults; P < 0.001, uncorrected, in children): compared with healthy adult subjects and control pediatric patients, long–COVID-19 patients showed extensive areas of hypometabolism including the orbitofrontal cortex bilaterally (in children at a liberal statistical threshold only), the medial temporal lobes bilaterally (hippocampus and amygdala; in adults right side only), the right thalamus (in adults only), and the brain stem and cerebellum. In adults, the number of complaints showed a weak negative association with brain stem and cerebellum uptake (r² = 0.1 and 0.34, respectively). In their initial study on adults, the authors proposed neurotropism of SARS-CoV-2 through the olfactory bulb, with extension of impairment to the limbic and other mentioned regions (43), whereas they provided additional alternative explanations in the latter study on children (e.g., inflammatory, dysimmune or vascular changes, disturbance of gut–brain relationship, or psychologic causes (44)). Sollini et al. (45) conducted a case-control study enrolling 13 patients with symptoms (mostly dyspnea and fatigue; Table 2) persisting longer than 30 d after infection recovery (average, 4.4 mo) and extracted brain scans from whole-body ¹⁸F-FDG PET/CT examinations. Compared with surgically treated melanoma patients, patients with long–COVID-19 exhibited hypometabolism of the right parahippocampal gyrus and thalamus at a liberal statistical threshold (P < 0.001, uncorrected). Exploratory analyses (P < 0.005, uncorrected) linked symptoms such as anosmia, ageusia, or fatigue to additional regions such as the orbitofrontal cortex or brain stem (substantia nigra), respectively.

In contrast, Dressing et al. (46) recruited patients with neurocognitive symptoms persisting for more than 3 mo after infection (6.6 mo on average; retrospectively, all meeting the criteria of the post–COVID-19 condition by the WHO (20)). All patients complained of attention and memory problems, most also complained of fatigue, and 39% could not return to previous levels of independence or employment. However, on extensive neuropsychologic testing, half the patients were completely unimpaired, whereas other patients showed mild to moderate impairment in single domains (most frequently visual memory, 21% of cases). The most consistent finding was fatigue (61%; cognitive fatigue, 67%) when using a self-rating questionnaire. Fourteen patients underwent ¹⁸F-FDG PET that yielded no significant regional abnormality in comparison to control patients on either a single-subject or group level (using both SPM [P < 0.005, uncorrected] and principal-components analysis; Fig. 1). No correlation between clinical variables and PET measures could be established. Given the striking discrepancy between these findings and findings in COVID-19–related encephalopathy, the authors proposed that mechanisms other than those in encephalopathy probably contribute to post–COVID-19 syndrome, fatigue being foremost (46).

Study Design and COVID-19 Populations

Given the still high incidence of COVID-19, there is no necessity to rely on retrospective analysis of convenient samples. They pose an inherent risk of bias and, because of the lack or inconsistency of data, do not allow for in-depth analyses of clinicoimaging correlations. Likewise, case reports or series are hardly justified except for very rare conditions, for which they may constitute an interesting starting point. In addition, it is evident that currently available longitudinal studies provided the deepest insight into underlying mechanisms and their course (39,41).

The latter studies demonstrated a strong time dependency of COVID-19–related CNS changes, underlining the need to clearly define the time window of inclusion of patients with respect to the time of SARS-CoV-2 infection. Likewise, it is apparent that a large fraction of inpatients has COVID-19 encephalopathy (∼70% (18)), which takes several months to recover from (39,41,48). Despite objective cognitive impairment in these patients, subjective perception of deficits and psychologic strain is frequently lacking in this group (49). In contrast, symptoms such as stress, anxiety, and reduced wellbeing are more prevalent after a mild course of disease (49). Accordingly, long-term symptoms such as subjective cognitive impairment and fatigue typically occur in patients who are younger (<50 y) and have a low rate of hospitalization (<30%; (2,19)). As the definition of “post–COVID-19 syndrome” used so far reflects only temporal aspects (5), there is a tremendous risk of building mixed populations in terms of initial disease severity, treatment, and impairment (subjective/objective). Proper selection of patients will be key in separating protracted COVID-19–related encephalopathy from the post–COVID-19 syndrome in a strict sense (2,19). In this respect, the novel, stricter WHO definition is welcome progress (20). In fact, available PET studies do not support the idea that neurocognitive post–COVID-19 syndrome represents a protracted manifestation of COVID-19–related encephalopathy, because available studies on long–COVID-19 or post–COVID-19 syndrome either yielded no metabolic alteration at all (46) or depicted metabolic changes (43–45) that do not match the highly consistent findings in COVID-19–related encephalopathy (18,39). Concerning the diagnosis of COVID-19, a PCR or at least state-of-the-art antibody proof should be mandatory. From a clinical perspective, typical clinical findings may be sufficient to ground a diagnosis in a pandemic situation, but this is not sufficient for a scientific study trying to unravel a specific disease.

Up to now, none of the reviewed studies has conducted a proper sample size calculation. Likewise, study registration appears to be an exception (e.g., (18,46)) that needs to be turned into a rule.

Control Populations

The use of convenient samples seems to be an even bigger problem when it comes to PET reference datasets. In fact, the descriptions of methods of most studies do not clearly state whether reference populations were indeed examined under comparable circumstances (e.g., identical scanners, acquisition techniques, reconstruction techniques, or patient preparation). The contribution of technical differences may be negligible if disease effects are large but may be a crucially biasing factor if effects are small (e.g., post–COVID-19 syndrome). Some authors acknowledge inconsistencies and conducted confirmatory analyses (18), whereas most authors ignored this issue. The use of oncologic control patients may be particularly problematic if brain scans are cropped out of whole-body scans, which are usually acquired and reconstructed differently from brain scans. Furthermore, preparation of brain scans requires well-defined resting (or neutral) conditions, whereas adherence to these conditions is lower in patients undergoing whole-body PET/CT (patients may read, talk, or use their smartphones).

Since the publication of the original studies by our group relying on control patients (i.e., patients with unclear neurologic or psychiatric complaints, in whom a somatic CNS disease was carefully excluded (18,41,46)), we had the opportunity to scan a cohort of thoroughly evaluated healthy controls. The local ethics commission and federal office for radiation protection approved this study, and all subjects gave written informed consent. Using these subjects as a reference cohort, we were able to confirm our initial results (Fig. 1). We also conducted an additional analysis on subjects with post–COVID-19 syndrome and hyposmia, but we found no significant hypometabolism of the olfactory regions as described by other groups.

PET Data Analysis

Comparison of datasets by common statistical methods such as SPM requires count rate or intensity normalization. It is well known that extensive, disease-related metabolic alterations in one direction may cause artificial alterations in the other direction on conventional analyses, if diseased brain regions are part of those used for normalization. In cases of neocortical hypometabolism and normalization by whole-brain proportional scaling, apparently “hypermetabolic” areas often include the basal ganglia, thalamus, mesiotemporal regions, brain stem, and cerebellum (40). Several lines of evidence such as ROI analyses (plasma glucose–corrected uptake values), clinicoimaging correlations, and follow-up studies suggest that the areas of hypometabolism in subacute settings most likely reflect true hypometabolism (40). However, in our opinion, and with the possible exception of concomitant autoimmune encephalitis, it is still questionable whether regions of hypermetabolism reported in several studies exhibit true hypermetabolism or inflammation. On one hand, the latter is not convincingly supported by histopathologic data. On the other, an artificial increase seems much more likely, especially if global count rate normalization was applied (40). Still, given the systemic activation of microglia and astrocytes (albeit being most pronounced in the brain stem and cerebellum), one cannot exclude that associated metabolic changes cause some bias if white matter is used as a reference region (18,41). One way to avoid this problem may be to conduct principal-components analyses as proposed by Hosp et al. (18).

Given the complexity, costs, and radiation exposure, cohorts in PET studies are often rather small, which limits their statistical power. Thus, a pragmatic approach may be to define a liberal statistical threshold and avoid a correction for multiple comparisons. This may lead to type I errors. Several systematic studies (Tables 1 and 2; Supplemental Table 2) used liberal thresholds; consequently, this limitation needs to be kept in mind when considering anatomically (e.g., small clusters near or within CSF spaces (45)) or functionally (e.g., weak correlations with subjective, ill-defined complaints (43)) questionable findings. Such limitations particularly apply to case reports or smaller series, which often suffer from additional flaws (e.g., reliance on observer-dependent visual reads, lack of coregistration to high-resolution structural imaging or MRI; Supplemental Table 1).

Clinicoimaging Correlations

Whereas clinicoimaging correlations cannot prove causality, they may be useful to check for plausibility. For instance, the high correlation of frontoparietal-dominant changes in cerebral metabolism and formally assessed cognitive functions served by frontoparietal regions in COVID-19 encephalopathy (18) supports the notion that these changes depend on each other. The magnitude of metabolic alterations was also in line with imaging studies on patients with comparable cognitive deficits (e.g., dementia).

In turn, some imaging findings need further elucidation: for instance, bilateral hyperactivation of the inferior colliculi (36,37) may be expected to interfere with normal auditory functioning, which, however, was not reported. Given that this finding was reported in rather severely affected patients, one may also consider alternative explanations (e.g., artificial, relative hypermetabolism due to neocortical hypometabolism). Donegani et al. (33) reported a bilateral hypometabolism of the parahippocampal and fusiform gyri in isolated hyposmia. Although they used an 8-item odor test, no correlation analyses were reported. In turn, one may also have expected other deficits such as memory impairment, which unfortunately was not addressed. In cases of post–COVID-19 syndrome, the situation is even more complex: patients commonly complain about multiple symptoms with a high impact on daily functioning, which, however, cannot be verified in such magnitude by formal testing (46,49). Still, extensive patterns of hypometabolism have been reported (43,44), which on first glance may be expected to result in various additional symptoms (e.g., severe objective amnestic problems). Unfortunately, no formal testing was pursued. In contrast, no such metabolic changes were observed in the study by Dressing et al. (46), underscoring the need to verify symptoms and complaints by operationalized formal assessments to gain deeper insights and unravel discrepancies. Such assessments will also benefit from inclusion of additional correlative biomarkers (e.g., PET or CSF biomarkers of neurodegeneration) and, ultimately, imaging–neuropathology correlations, which are still very sparse (18).

Correlations with Knowledge from Related Disciplines

Imaging findings and concepts derived from them have to be aligned with knowledge from related disciplines. It is evident that neurologic or neurocognitive findings in COVID-19 cover a broad spectrum of disciplines. Neuropathologic findings and psychiatric concepts may be considered as 2 distant ends of such a spectrum. However, we believe that these disciplines are particularly helpful for a better understanding of current findings.