Magnetic Resonance Imaging of Cerebral Aspergillosis: Imaging and Pathological Correlations

Cerebral aspergillosis is associated with a significant morbidity and mortality rate. The imaging data present different patterns and no full consensus exists on typical imaging characteristics of the cerebral lesions. We reviewed MRI findings in 21 patients with cerebral aspergillosis and correlated them to the immune status of the patients and to neuropathological findings when tissue was available. The lesions were characterized by their number, topography, and MRI signal. Dissemination to the brain resulted from direct spread from paranasal sinuses in 8 patients, 6 of them being immunocompetent. Hematogenous dissemination was observed in 13 patients, all were immunosuppressed. In this later group we identified a total of 329 parenchymal abscesses involving the whole brain with a predilection for the corticomedullary junction. More than half the patients had a corpus callosum lesion. Hemorrhagic lesions accounted for 13% and contrast enhancement was observed in 61% of the lesions. Patients with hematogenous dissemination were younger (p = 0.003), had more intracranial lesions (p = 0.0004) and had a higher 12-week mortality rate (p = 0.046) than patients with direct spread from paranasal sinuses. Analysis of 12 aneurysms allowed us to highlight two distinct situations. In case of direct spread from the paranasal sinuses, aneurysms are saccular and located on the proximal artery portions, while the hematogenous dissemination in immunocompromised patients is more frequently associated with distal and fusiform aneurysms. MRI is the exam of choice for cerebral aspergillosis. Number and type of lesions are different according to the mode of dissemination of the infection.


Introduction
Aspergillosis is a filamentous fungal infection transmitted by inhalation of airborne spores or by contamination of wounds [1]. This disease develops primarily in immunocompromised patients and may progress to disseminated aspergillosis defined by the infection of at least two noncontiguous organs. The brain is one of the most commonly affected secondary sites and is involved in 10-44% of patients [2][3][4][5][6][7][8]. Cerebral aspergillosis is the worst clinical form with a 85-100% mortality rate [9,10]. Despite the development of Aspergillus galactomannan and the beta-D-glucan detection tests as well as PCR, the mycological data are not always conclusive and imaging modalities such as CT-scan for invasive pulmonary aspergillosis (IPA) and CT combined with MRI for sinus and cerebral aspergillosis are crucial. A review of literature on the MRI appearances of cerebral aspergillosis shows that a few lesions are enhanced after gadolinium injection with a nodular or annular presentation (Table 1) [4][5][6][11][12][13][14][15][16][17][18][19][20][21][22][23]. As for other abscessed lesions, the most common pattern consists of a T1W hypointense signal and a T2W hyperintense signal. However, in the literature some presentations stand out, including mainly a T2 intermediate with surrounding high signal, a T1 hyperintense ring, and a target-like ADC (apparent diffusion coefficient) signal [4,12]. A few cases of mycotic aneurysms due to Aspergillus have also been reported [17,20]. The purpose of this study was to describe the variety of MRI presentation of cerebral aspergillosis precisely and to compare these presentations to pathological findings.

Patients and Methods
We retrospectively reviewed neuropathological, clinical and neuroimaging findings of 21 patients with cerebral aspergillosis. These cases were identified over a period of 8 years (from 2006 to 2013) and collected in seven French and Belgian university hospitals. The study was approved as a retrospective non-interventional study by the local ethics committee (Comité d'Ethique de la Faculté de Médecine, d'Odontologie, de Pharmacie, des Ecoles d'Infirmières, de Kinésithérapie, de Maïeutique et des Hôpitaux, Strasbourg). Two trained neuroradiologists analyzed imaging data to describe the radiological semiology of the different lesions. The results were compared to the patient's pathological findings and immune status, in cooperation with a neuropathologist and a specialist in clinical mycology. The Fisher exact test and Mann-Whitney test were performed when appropriate using GraphPad Prism version 6.00 (Graph-Pad Software, La Jolla, CA). Diagnosis of invasive aspergillosis was based microscopy and culture of respiratory samples, microscopy, culture and histopathology of sinus, brain or other tissue biopsies or on positive Aspergillus galactomannan detection test in serum, CSF or bronchoalveolar lavage fluid. Tissues were stained by periodic acid-Schiff, hematoxylin-eosin and Grocott's methenamine silver stain. Tissue samples came from surgical biopsy (n = 9) or necropsy (n = 6). For each case, the fungal infection by Aspergillus was classified as proven or probable according to a strict application of the 2008 EORTC/MSG definitions [24]. One patient (#11) had initially an IPA with a rapid dissemination to the brain documented by clinical signs and MRI and secondary presented also a Trichosporon mycotoxinivorans fungemia and ultimately an additional Scedosporium apiospermum disseminated infection [25]. This case was included in the study because the cerebral infection was documented while A. fumigatus  [12,26].

Results
Twenty-one cases appropriately explored by MRI were identified. Most patients were immunosuppressed (15/21, 71%), due to organ transplantation treatment for seven of them. A majority of the immunocompromised patients had primarily IPA (11/15). The patients' sex, age, risk factors, immunological status, mycological positive findings, degree of certainty of the aspergillosis, organs involved, cerebral lesion characteristics, and outcome are detailed in Table 2.

Direct spread from paranasal sinuses
In eight patients, the cerebral infection occurred during sinusitis (n = 6) or after a craniofacial trauma (n = 2). Only two of them were immunocompromised, the other six being immunocompetent. We identified nine parenchymal lesions. Five patients had a single lesion. T2 Ã W imaging showed a significant peripheral low signal with susceptibility artifacts in favor of bleeding complications (n = 8). DWI highlighted the presence of one target-like ADC signal lesion in two patients (Table 3). MRI typically demonstrated a peripheral invasive lesion starting next to the nasosinusal cavities. Its outlines were polylobulated with a well-defined rim enhancement after gadolinium. The spreading by contiguity involved the concentric layers of the CNS from the dura mater to the white matter accompanied by a sizeable inflammatory phenomenon and a significant vasogenic edema (Fig 1). The progress of the abscess into the brain was accompanied by a subdural empyema in four patients. All patients underwent surgery and survived the infection with a follow-up of 7 months to 5 years.

Hematogenous dissemination
Thirteen patients had a total of 329 parenchymal lesions ( Table 4). All of these patients were immunosuppressed. The infection involved all location in the brain with a predilection for the corticomedullary junction in the supratentorial area (203/329, 62%), white mater (14%), basal nuclei-thalamus (9%) and infratentorial structures (7%). A corpus callosum lesion was seen in six patients. A hemorrhagic component was seen on T2 Ã -weighted images as a significant hypointense signal in 42 lesions (13%). We noted contrast enhancing for 168 lesions (61%), with a more frequent annular shape (n = 96) compared to the nodular type (n = 72). A T2W intermediate signal with a surrounding high signal appeared for 26 lesions (8%) and 16 lesions for the target-like ADC signal (5%), with a minimum size of 9 mm (Figs 2 and 3). Comparison according to the dissemination pathway showed that patients with hematogenous dissemination were younger (p = 0.003), more often had immunosuppression (p = 0.005), had more intracranial lesions (p = 0.0004) and had a higher 12-week mortality rate (p = 0.046) than patients with direct spread.  Table 5 summarizes the main points concerning the vascular lesions. Two immunocompetent patients with direct spread presented macroscopic vascular complications. We identified three saccular arterial aneurysms with a proximal topography (internal carotid, sphenoidal segment of the middle cerebral artery, anterior communicating artery). Moreover, one patient developed a cerebral venous sinus thrombosis. Two immunocompromised patients with hematogenous dissemination had a total of nine aneurysms: six distal fusiform and three proximal aneurysms (two saccular and one fusiform). The death of these two patients was due to aneurysmal rupture causing a massive cerebral hemorrhage. Furthermore, the aneurysm of the distal part of the basilar artery causing death of patient #13 was not visualized on the imaging studies (CT, cerebral angiography) (Fig 4). This phenomenon could be explained by an aneurysmal collapse during the acute phase.

Gross pathology examination findings
Postmortem examinations were only performed on patients with hematogenous dissemination. These procedures revealed the presence of disseminated abscesses involving the whole brain with a predilection for the corticomedullary junction. In some cases, the autopsy identified hemorrhagic complications such as intralesional bleeding or subarachnoid hemorrhage. Arterial dissection revealed the presence of both proximal saccular and distal fusiform aneurysms.

Microscopic examination findings
The microscopic analysis of the vascular elements revealed that some vessels had been thrombosed by fungal invasion. Infected vessels showed wall destruction by hyphal invasion associated with an area of fibrinoid necrosis leading to red blood cell extravasation (Fig 5A and 5B). The aneurysmal wall was characterized by interruption of the internal elastic lamina and advanced destruction in which identification of the wall components (intima and media) was no longer possible. The pathologic wall was invaded by hyphae and inflammatory cells (granulocytes, giant cells) (Fig 5C and 5D). The most frequent lesions identified were abscesses. The microscopic examination highlighted three-layer lesions with a necrotic center, surrounded by a dense hyphal rim encircled by granulation tissue with or without giant cells (Fig 6). The abundance of granulocytes with or without giant cells is usually inversely related to the density of hyphae.

MRI evolution in a long-term survivor
Data from patient #10 who had sequential MRI assessments revealed persistent contrast enhancement after 30 months for the largest lesion (thalamic topography initially measuring 2.6 cm) (Fig 7). Originally, the enhancement was annular and progressed towards a nodular shape after 3 months. A majority of lesions disappeared after 6 months on DWI, which seems to be the most sensitive sequence. The lesions increased in number and size on MRI performed 2 weeks after start of voriconazole therapy. The smallest lesions evolved to complete disappearance, leaving a hypersignal on the FLAIR sequence. One of the corticomedullary junction lesions evolved to a punctiform calcification in a granulomatous process. The progression of the target-like ADC signal was marked by the disappearance of a central hypointense signal on DWI. For the largest thalamic lesion, we also noted a persistent DWI hypersignal over 1.5 years.

Outcome
Eight (38%) of the patients died within 12 weeks. All were immunosuppressed and died of aspergillosis. Two immunosuppressed patients died beyond week 12, one of aspergillosis and one of another cause, 11 and 16 months after diagnosis respectively.

Discussion
Cerebral aspergillosis has a poor prognosis. In our literature review on MRI aspects in cerebral aspergillosis (Table 1), based on studies published between 1994 and 2013, the mortality rate was 86%. In our patients, the 12-week mortality rate was surprisingly low (38%) with a significant difference in the six immunocompetent patients (0%) and in the 15 immunosuppressed patients (53%). Possible explanations for the improvement of survival in our series are the treatment with a voriconazole containing antifungal regimen, the use of surgery in nearly half of the patients and lack of immunosuppression in six (29%) of the patients. Voriconazole is now the gold standard for the treatment of IPA [27]. Its activity in cerebral aspergillosis has been demonstrated with only a death due to aspergillosis in only 37 patients (46%) out of 81 patients treated with voriconazole [28]. The critical role of surgery is also suggested in this study. Comparing different MRI sequences (including T1W, T2W, FLAIR, T2 Ã W, T1W after gadolinium, and DWI), the investigations confirm that diffusion-weighted imaging appears to be the most sensitive modality for early identification of cerebral aspergillosis [12,13,16,18]. Direct spread of infection essentially begins from the paranasal sinuses. In accordance with    the literature, a majority of the patients were immunocompetent [5,17,20]. The presentation is fairly similar between the different cases. MRI reveals a polylobulated abscess with peripheral contrast enhancement, substantial inflammation involving adjacent structures (paranasal sinuses, dura mater with focal meningitis, osteomyelitis) and a significant parenchymal edema. Subdural empyema may be associated. For the immunocompromised patients, cerebral invasion most commonly results from hematogenous spread. Histological investigations confirm the accepted mechanism consisting of an arterial occlusion by a septic thrombosis. The fungus passes through the wall of the blood vessel by producing elastase, which digests the internal elastic lamina and causes abscess formation. We observed distinctive lesions displaying central coagulative necrosis surrounded by a dense hyphal rim and a more external granulocyte layer. This physiopathology explains the topographic specificities. Indeed, lesions have essentially been seen at the corticomedullary junction, white matter, basal nuclei and thalami. These data can also be attributed to the vascular infection mode involving cortical arteries, which may be the first to lose their patency because of their narrower diameter [13]. The impairment of the basal nuclei and thalami is also characteristic and indicates the involvement of lenticulostriate and thalamoperforator arteries [5,6,11]. In our study, more than half of the patients had a corpus callosum lesion. This structure is involved in relatively few other processes (e.g., high-grade astrocytoma, cerebral lymphoma, multiple sclerosis and Marchiafava-Bignami disease), which can be readily differentiated from aspergillosis on the basis of their clinical and biological context [13]. Prior studies have assumed that contrast enhancement after gadolinium injection is correlated to the immune status in the sense that profoundly immunocompromised patients have no enhancement because of the absence of inflammatory response [4,6,12,21,29]. It is also accepted that a well-defined ring enhancement is quite uncommon [29]. Moreover, several researchers have suggested that the absence of enhancement seems to be a poor prognosis factor [6,23]. In our literature review, for 65 patients from comparable studies, 59 were immunosuppressed. Contrast enhancement was observed in approximately 55-60% of the patients [4][5][6][12][13][14][15]18,22]. The analysis of our data confirms that 61% of the lesions were enhanced in both the entire group and the immunocompromised group. Contrast enhancement is not correlated to the immune status because many profoundly immunosuppressed patients had well-defined enhancing lesions. In addition, none of our data suggest that the absence of contrast enhancement is associated with a higher 12-week mortality rate since half of the deceased patients showed enhancement (4/8 patients). Analysis of the T2-and T2 Ã -weighted hypointense signal is difficult (Fig 8). Previous articles had addressed the issue, explaining that this signal could be the result of either ferromagnetic fungal deposits (iron, manganese, magnesium, zinc) [30] or the presence of methemoglobin (in the capsule wall and/or the macrophages) or the presence of free radicals produced by the macrophages [5]. Histologically, a mild T2 Ã -weighted hypointense signal does not appear to be associated with a significant hemorrhagic phenomenon, although some red blood cells are associated with a significant amount of hyphal structures in this area. It is therefore difficult to separate these phenomena. To check whether a lesion is hemorrhagic, we had to rely on a qualitative criterion such as a significantly high hypointense signal. According to this, an estimated 13% of lesions in our study were hemorrhagic, in comparison to the literature data reporting a 25-39% rate [14,23]. Two signal features on MRI could be more specific of a cerebral Aspergillus abscess. Firstly, Miaux et al. described a central intermediate signal on T2-weighted sequences, corresponding to coagulative fungal necrosis at autopsy findings [4,12]. This feature is surrounded by a high hyperintense signal corresponding to vasogenic  edema and accounted for 8% of the total lesions. Secondly, the present investigation confirms the presence of target-like ADC signal lesions, accounting for approximately 5% of the lesions. In immunocompromised patients, ring lesions on DWI occur mainly in cerebral aspergillosis but also in progressive multifocal leukoencephalopathy and CNS lymphoma [31]. We could juxtapose this target-like imaging element to the histological findings and particularly the characteristic three-layer lesion, which could explain the signal modification (Figs 2 and 6). Central necrosis presents two components responding differently on DWI. This feature is probably due to a difference in density impacting the Brownian motion of water molecules. The intermediate layers composed of a dense hyphal rim and peripheral inflammatory hypercellularity are visualized by contrast enhancement and a mild hypointense signal on magnetic susceptibility sequences (T2 Ã W) but has no match on DWI because of an insufficient spatial resolution. Mycotic aneurysms historically have accounted for 2-4% of all intracerebral aneurysms. Aspergillus is the most common agent that causes fungal aneurysms. It is considered that these lesions have a high risk of rupture [32]. Our results suggest that the two spreading pathways should be distinguished. The direct spread from paranasal sinuses produces saccular aneurysms on the proximal arterial portion from the circle of Willis. This process starts from the external vessel surface, penetrates the vasa vasorum system and gradually affects all layers including the adventitia, which leads to the weakening of the wall and subsequent aneurysm formation [17,20,33]. Conversely, the hematogenous dissemination in immunocompromised patients more frequently leads to fusiform aneurysms with a distal topography. In this context and compared to the literature data, our two patients (patients #11 and #13) are rare examples of fungal aneurysms without sino-nasal involvement. The follow-up of a long-term survivor has raised some interesting points. Firstly, this patient had the highest number of lesions, which could suggest that the extent of the disease may not be correlated to mortality. The only poor prognosis factor appears to be advanced immunosuppression. Secondly, the largest residual cerebral lesion, located in the thalamus, was still enhancing 2.5 years after onset of the disease. Such long-term contrast enhancement has already been described by Adler et al. regarding a basal ganglia fungal lesion on an enhanced MRI study obtained 3 weeks after the onset of neurologic symptoms [4,34]. Thirdly, the comparison of the first three MRI performed showed that lesions may increase in number and size even after 2 weeks of effective treatment with voriconazole. Fourthly, the initial examinations show annular contrast enhancement for almost all lesions. Progression will be towards nodular enhancement as the lesion regresses. In general, the progression of the disease is marked by the complete disappearance of the smallest lesions leading to a residual FLAIR hypersignal relating to a gliosis phenomenon. For the majority of lesions, the DWI hyperintense signal remains present for approximately 6 months. However, we have noted the presence of a lesion evolving like a granulomatous structure leaving a calcified scar. On the necropsy from another patient, a giant-cell granuloma was identified and could explain this type of imaging course. This last issue has been reported by Ashdown et al., among others, stating that calcifications likely represent healing or healed granulomas and are seen more often if the patient survives the acute infection [5,13,29].

Conclusion
Cerebral invasion during aspergillosis is a severe and often fatal complication. To increase the chances of survival, physicians need to diagnose infection earlier to introduce rapidly an effective antifungal treatment. MRI appears to be the exam of choice and should require at least T1W, T2W, FLAIR, T2 Ã , DWI and T1W after gadolinium injection sequences for diagnosis and follow-up.