We showed brain postmortem high-resolution images of COVID-19-deceased patients, with histological correlations.
Cerebrovascular abnormalities were the most common findings, detected in five patients: one stroke (although timing was not established), one subarachnoid hemorrhage (although she had also undergone thrombolytic therapy for pulmonary embolism) and three patients with microhemorrhage. They are the most commonly described CNS imaging finding in COVID-19 patients , and potential mechanisms include endothelial damage linked to the angiotensin-converting enzyme 2 (ACE2) receptor, hypercoagulable state and cardiomyopathy .
A recent meta-analysis of 108,571 patients with COVID-19  showed that acute cerebrovascular disease occurred in 1.4%, with acute ischemic stroke accounting for 88% of those and intracerebral hemorrhage for 12%. They were more frequent in older subjects, in those who were severely infected and had preexisting vascular risk factors. Compared to individuals who experienced a stroke without the infection, patients with COVID-19 and stroke were younger, had higher National Institutes of Health Stroke Scale/Score (NIHSS), higher frequency of large vessel occlusion and higher in-hospital mortality rate. The pattern of large vessel occlusion and multi-territory infarcts suggests that cerebral thrombosis and/or thromboembolism could be possible causative pathways for the disease. Hemorrhage could be associated with arterial wall damage and rupture related to SARS-CoV-2 affinity for ACE2 receptors, which are expressed in brain endothelial and arterial smooth muscle cells. Alternatively, hemorrhagic infarcts could be related to the inflammatory response .
Intra-axial susceptibility abnormalities suggestive of microhemorrhage were the most common finding (74%) in a consecutive cohort of COVID-19 patients  and have also been reported in postmortem brain analysis in COVID-19 patients . Nonetheless, these abnormalities have also been described in critically ill patients, patients using extracorporeal membrane oxygenation (ECMO) and in high-altitude hypoxia, and are not, therefore, specific to COVID-19. In our samples, microvascular brain damage was detected in three cases on MRI (and in 5 cases in pathology, 3 of them being the ones with MRI findings). Pathological analysis was superior to imaging in detecting perivascular hemorrhages, possibly due to higher spatial resolution (pathologically detected microhemorrhages were very small and adjacent to the vessel, and together, they were smaller than the in-plane resolution of the SWI acquisition (0.20 mm), making it impossible to differentiate one another: For example, in Fig. 6e there is hemosiderin adjacent to the vessel, and together, they measure 0.15 mm; in Fig. 6f, the vessel and adjacent hemorrhage measure 0.10 mm). Besides this issue, the SWI artifact related to blood products is usually bigger than the blood products themselves, being another explanation why these small perivascular hemorrhages could not be differentiated from the vessels in the MRI. This is particularly a problem in the in situ postmortem brain 7T MRI, because the intravessel susceptibility signal (from postmortem intravessel stasis/thrombi) is very strong in this scenario, as shown in Fig. 4d for example. Although a previous paper showed underestimation of CMBs by postmortem MRI brain compared to histology, in up to 24%, another study using 7T showed better correlation with MRI, especially for those in the cortico-subcortical regions (using ex situ images, which have much smaller intravessel SWI, because much of the blood has been washed away during fixation). Another issue is that, besides the resolution, MRI might also show false positives. The same study using 7T MRI and postmortem histology showed that in the deep white matter 42% of microhemorrhages were not detected, while 31% of T2* hyposignals were not due to microhemorrhages, but due to vessels filled with postmortem thrombi , adding even more nuances to this diagnosis.
One patient (patient 2) had other imaging findings besides hemorrhage. Posterior cortical–subcortical abnormalities could represent posterior reversible encephalopathy (PRES), which have been described in COVID-19-deceased patients , and possibly related to endothelial cell damage due to viral ACE receptor binding. Clinical features were also compatible with PRES, that is commonly associated with seizures. An alternative explanation for these imaging findings would be hypoxic changes. Imaging findings in the basal ganglia could represent PRES, encephalitis, metabolic-related changes or hypoxic injury. One of the sampled fragments included part of the basal ganglia. Histological analysis of that fragment did not disclose any evidence of encephalitis, such as perivascular infiltrates, glial nodules or cytopathic changes. Most of the histological changes in this case appear to represent systemic illness responses and hypoxic injury. Hippocampal changes could also be related to hypoxia. Clinical history corroborates hypoxic injury, since, in addition to prolonged mechanical ventilation related to hypoxia, the patient also suffered four episodes of cardiorespiratory arrest during hospitalization. Corticospinal tract signal abnormalities have not been previously described in COVID-19 patients, although paraparesis has been reported. Wallerian degeneration resulting from cortical hypoxia is a possible explanation. These findings were supported by axonal thickening and disruption on Bielschowsky stain. The axonal lesion (so-called Wallerian-like degeneration) is well described and associated with ischemic lesion in experimental and clinical studies [15, 16]. Although it can be difficult to distinguish from traumatic brain lesion, the other histopathological changes and the clinical context can help in differential evaluation [16, 17]. A direct point-to-point correlation between imaging and histology findings was not possible, limiting interpretation of imaging and pathology findings.
A focal lesion in the splenium of the corpus callosum was identified in patient 4, which may occur in diverse clinical scenarios, such as epilepsy, encephalitis and drug-related, and has been described in COVID-19 patients , especially in children with MIS-C (multisystem inflammatory syndrome in children) . These lesions are usually transitory and are characterized by increased T2 and FLAIR signal, with diffusion restriction. Some authors propose that cytokine‐mediated edema may underlie this finding  which would be in agreement with this prevalence in children with MIS-C. Hemorrhagic foci, such as seen in this case, are not frequently associated with these transitory lesions, but have been previously described in COVD-19 patients .
Finally, white matter hyperintense lesions were detected in two patients (Patients 3 and 5). The lesions usually represent a nonspecific finding, which may be related to chronic white matter small vessel disease. Acute white matter abnormalities have been reported in COVID-19 patients , especially centrum semi-ovale lesions showing restricted diffusion, sometimes associated with globus pallidus lesions , representing either ADEM-like lesions or, most likely, ischemic/vasculitic lesions [22, 23]. In one case, tissue sampling was not performed, and in the other one, histological analysis showed chronic small vessel changes. Small vessel disease is a common finding in patients with hypertension and diabetes, underscoring the relevance of preexisting conditions as risk factors for an unfavorable outcome.
Previously reported imaging findings in COVID-19 patients, such as meningeal enhancement (that cannot be adequately assessed by postmortem studies), gyral swelling and diffuse white matter signal abnormalities were not detected in our series .
Regarding the clinical scenario, all patients were critically ill, in the intensive care unit, with respiratory failure requiring mechanical ventilation, associated with renal failure evolving to refractory shock. Headache, a commonly described complaint, was reported by three subjects as initial symptoms. One patient, who showed the most prominent abnormal imaging finding, presented with new onset seizures leading to hospital admission.
Of all the patients in our series, only one underwent in vivo brain imaging (normal head CT). This is a limitation of this study. Corresponding in vivo imaging would be ideal to better differentiate postmortem aspects from in vivo findings. Typical postmortem brain findings and challenges include: different T1, T2 and ADC due to a combination of factors like lower temperature and decomposition (that can be affected by postmortem interval, ambient, etc.); loss of grey–white matter differentiation mainly on CT, but with some blurring on MRI, due to fluid shift and cerebral autolysis; drastically different susceptibility-weighted image due to a darker vessel signal mainly related to blood stasis (which is even more exacerbated in 7T imaging); gas appearing linked to putrefaction; no intravenous contrast administration (although postmortem angiography might be done—in which case brain enhancement might be normally expected) ; hyperdensity (or clot signal on MRI) of the dural venous sinuses, sometimes in a dependent position; increase in brain volume and loss of definition of sulci, which may be due to a combination of vasogenic and cytotoxic edema and may be more pronounced in death with prolonged duration of the agonal state, leading to a longer hypoxic state, than in acute death [25, 26]; and in more advanced deterioration “softening” of brain tissue and settling of the tissue against the dependent part of the skull [27,28,29,30]. Obtaining in vivo neuroimaging, especially MRI, in critically ill COVID-19 patients, with high infectious potential is limited, leading to underestimation of neurological involvement in these patients.
The pathological analysis of our samples showed reactive gliosis, congestion and cortical neuron eosinophilic degeneration, axonal disruption and the presence of SARS-CoV-2 antigens in the endothelial cells and microglia and RNA detection in all analyzed cases. Edema was identified in 5 cases (perivascular scarce neuropil, forming “vacuoles”), exudation of fibrin and hemosiderin (meaning that an extravascular leak of red cells has occurred, days before the death) (Fig. 6c–f). This is a most probably diffuse multifactorial pathological process, involving the endothelial cells, leading to extravascular leakage: SARS-CoV-2 infection of endothelial cells, as we show positive detection of the N-antigen in the cytoplasms of those cells (Fig. 6j, l); hypoxia and shock. Alzheimer type II astrocyte was identified in 3 patients. It represents a common change in (and an important marker for) patients with metabolic impairments, notably renal and hepatic failure [31, 32]. Although their origin and relation to clinical manifestations is unclear, some studies (specially experimental ones) propose that these cells arise in the same pathophysiological background as the neurological manifestations (changes in glutamine and ammonia levels) . Only one case showed mixed periventricular inflammatory infiltrates (Patient 7). Neuropathological studies on COVID-19 are scarce so far. One case report disclosed findings suggestive of both vascular and demyelinating processes . Another study included 18 brains from deceased COVID-19 patients  and showed mainly hypoxic injury in the cerebrum and cerebellum, with neuronal loss in the cerebral cortex, hippocampus, and cerebellar Purkinje cell layer. These findings are concordant with ours. In two cases, rare foci of perivascular lymphocytes were detected, and leptomeningeal inflammation was detected in one brain specimen. Similar findings were seen in brains from 43 COVID-19 fatal cases . The commonest finding was reactive microglia (86%), with inflammatory infiltrate by T cytotoxic cells, mainly in the brainstem, cerebellum and leptomeninges. They also detected SARS-CoV-2 in the brain of 53% of the patients, and inflammatory reaction was not associated with SARS-CoV-2 immunohistochemistry positivity (using anti-nucleocapsid protein and anti-spike protein primary antibodies) in the brain. Possible explanations would be that SARS-CoV-2 could infect specific central nervous system compartments; infection may not be associated with local inflammatory reaction, in an immune evasion mechanism, or producing different types of central nervous system involvement . Our series corroborates previous neuropathological studies that disclosed hypoxic injury , alterations related to the critical multiorgan failure condition of these patients, such as hemorrhagic foci and splenial lesions, as well as SARS-CoV-2 brain invasion, detected by immunohistochemistry and/or RNA detection.
We can highlight that IHC reactions were positive even in the samplings of brain with normal postmortem MRI.
Pathological–imaging correlation was limited due to the features of postmortem transethmoidal tissue sampling, since it was not possible to conduct a guided sampling of the representative area in a real-time correlation between imaging findings and anatomopathological examination. Thus, some imaging findings may not be represented in pathological samples. However, imaging and pathological findings complemented each other, since MRI included the whole brain (albeit with limited direct viral analysis, spatial resolution, etc.), while pathology was able to examine microscopic details, as well as IH and RT-PCR. In addition, these correlation links in vivo findings (which can be achieved with imaging studies), with pathological analysis (that are only possible with biopsies or postmortem), thus helping establish a better understanding of the in vivo imaging findings.
The sample size of the study was small, so more extensive studies are needed to have a better understanding of the overall neuropathological burden. Future directions include different tissue sampling approaches to try to address all imaging findings in postmortem examination in the autopsy room, additive histological techniques (e.g., electron microscopy and immunofluorescence), as well as extensive brain analysis of patients with different forms and moments of the disease.