Microbleeds show a characteristic distribution in cerebral fat embolism

This systematic review aims to test the hypothesis that microbleeds detected by MRI are common and show a characteristic pattern in cerebral fat embolism (CFE). Eighty-four papers involving 140 CFE patients were eligible for this review based on a systematic literature search up to 31 January 2020. An additional case was added from hospital records. Patient data were individually scrutinised to extract epidemiological, clinical and imaging variables. Characteristic CFE microbleed pattern resembling a “walnut kernel” was defined as punctuate hypointensities of monotonous size, diffusely located in the subcortical white matter, the internal capsule and the corpus callosum, with mostly spared corona radiata and non-subcortical centrum semiovale, detected by susceptibility- or T2* weighted imaging. The presence rate of this pattern and other, previously described MRI markers of CFE such as the starfield pattern and further diffusion abnormalities were recorded and statistically compared. The presence rate of microbleeds of any pattern, the “walnut kernel microbleed pattern”, diffusion abnormality of any pattern, the starfield pattern, and cytotoxic edema in the corpus callosum was found to be 98.11%, 89.74%, 97.64%, 68.5%, and 77.27% respectively. The presence rate between the walnut kernel and the starfield pattern was significantly (p < 0.05) different. Microbleeds are common and mostly occur in a characteristic pattern resembling a “walnut kernel” in the CFE MRI literature. Microbleeds of this pattern in SWI or T2* MRI, along with the starfield pattern in diffusion imaging appear to be the most important imaging markers of CFE and may aid the diagnosis in clinically equivocal cases.


Introduction
Cerebral fat embolism (CFE) is part of the fat embolism syndrome (FES), which results from the intravascular embolisation of fat globules. FES is most commonly caused by displaced long bone fractures, or during orthopaedic procedures but it may also be caused by other conditions like sickle cell disease, severe pancreatitis, or following liposuction [1,2].
First described by Zenker in 1862 the exact mechanism of CFE remains poorly understood. Both mechanical and biochemical theories have been proposed to explain the syndrome [3,4]. The mechanical obstruction of arterial circulation by neutral fat globules, followed by a delayed biochemical toxic injury caused by free fatty acids is the most probable mechanism of tissue injury [5,6]. CFE can occur without a patent foramen ovale, as small

Open Access
Insights into Imaging *Correspondence: giyab.omar@pte.hu 1 Department of Medical Imaging, University of Pécs Medical School, Ifjúság út 13, 7624 Pécs, Hungary Full list of author information is available at the end of the article Giyab et al. Insights Imaging (2021) 12:42 deformable fat globules can pass through the capillary circulation of the lung [7]. Cerebral fat embolism syndrome (CFES) occurs in up to 60% of patients with FES and is usually considered to be a self-limiting entity, but recent studies have also shown possible links to long term neurocognitive impairment [8][9][10]. The clinical presentation of CFES may vary considerably, symptoms typically appear between 12 to 72 h after the injury and may range from mild neurological impairment to alarming symptoms like coma or even death in the most severe cases [11].
There are no definitive clinical diagnostic tests and criteria developed, making the diagnosis of FES difficult [12,13]. The diagnosis is usually based on a combination of symptoms, laboratory and imaging findings [9]. Currently the most commonly used diagnostic criteria have been proposed by Gurd and Wilson presented in Table 1 [14]. In this system, at least two positive major criteria, or one positive major and four positive minor criteria are considered suggestive of FES [14].
Many patients, however, do not develop the classic triad of cutaneous, respiratory and neurological symptoms [9]. Accordingly, the Gurd and Wilson's criteria are not fulfilled, and the sole or first manifestation of FES may be neurologic, thus resulting in a diagnostic challenge to clinicians [9,15]. Post-mortem pathological studies have shown CFE to be more common than previously thought indicating that it is underdiagnosed, and the clinical criteria used to diagnose the condition are not entirely reliable [16].
CT scans performed are typically without any specific finding, however may show widespread interval low density changes which should raise the suspicion of CFE [11]. Compared to CT, imaging findings in MRI are much more common in CFE [17]. The "starfield pattern" has been widely used in literature for describing the reversible imaging pattern of multifocal punctate lesions showing diffusion restriction. Other reported features include confluent cytotoxic edema in the white matter, and vasogenic lesions that may enhance [1,11,17]. Additional features of CFE that has been increasingly reported more recently are cerebral and cerebellar microbleeds depicted by susceptibility weighted imaging (SWI) or T2* MRI [18]. These studies and a case that we present in this review have led us hypothesize that microbleeds show a characteristic pattern in CFE. We aimed to test this hypothesis by a systematic review of the literature.

Literature search
We have searched the PubMed database up to 31 January 2020 (no time constraints were set for the earliest publication) for papers which included detailed imaging description or images of cases with CFE. We used the keywords "MR" or "MRI" and"fat emboli" or"fat embolism", or "starfield pattern". References of the included papers were then individually checked for additional articles. We also searched the PACS, and the hospital information system (HIS) of the University of Pécs for CFE cases. We used the ICD-10-CM diagnosis code T79.10 (traumatic fat embolism) for searching the PACS (Med-View -ASPYRA LLC 7400 Baymeadows Way, Suite 101 Jacksonville, FL 32,256), and HIS (eMedSolution-T-Systems Magyarország Zrt. H-1097 Budapest, Könyves Kálmán krt. 36.) of the University of Pécs from 1 October 2016 (installation of MR scanner) until 31 January 2020. We have found one potentially suitable case for inclusion in this systematic review, whose legal representative was contacted and a signed informed consent was acquired for the anonymized inclusion and publication in this review. Search results were saved and organised via the commercially available Mendeley software (V1.19.5) and its web importer plug-in.
The exclusion criteria included recorded head trauma, poor image quality, unavailable images and poor image description, non-English language articles, animal experiment articles, CFE articles related to facial lipid injection procedure, and articles that are not available for online access.

Data extraction
All included articles were individually scrutinised by a board certified neuroradiologist (O.G.) and other two board certified radiologists with a minimum of five-year experience in interpreting neuroradiological examinations (A.T. and P.B.) to collect relevant data for this review. Data were populated into an Excel (ver. 1908, If more than one case was presented in an article, then all the available information from each case was recorded separately. If only mean values were reported for a given variable regarding patients, then these data were not included in further evaluation. The time point of CT and MRI scans were recorded in days, and time point 0 was considered to be the time of arrival to the hospital. If mentioned, the Glasgow coma scale (GCS) was used to record the neurological status. Gurd and Wilson's, and Schönfeld's criteria were only recorded if they were mentioned in the article.
The extracted clinical and epidemiological variables and available patients, are presented in Table 2.
Regarding imaging findings, the presented images, text and supplementary materials were evaluated as well. Imaging findings were recorded if they were identifiable on the presented images, or clearly described in the text. Findings that were not identifiable in the presented images or were not mentioned in the text were considered to be absent. In case of any discrepancy between the text and the available images, they were double-checked by each participating radiologist, and were excluded from the statistical analyses if the discrepancy was confirmed by all readers. Ambiguous imaging findings (due to poor image quality, or partially visible pathology) if not detailed in the text were not considered in the statistical calculations.
The hypothesized, CFE characteristic microbleed pattern was defined as the diffuse presence of round microbleeds (punctate focal hypointensities) of monotonous size in the subcortical white matter (involving but not limited to the U-fibers), internal capsule and the corpus callosum, mostly sparing the corona radiata and the non-subcortical centrum semiovale on T2* GRE or SWI images. As authors believe this pattern resembling the appearance of a walnut kernel, it is further referred to as the "walnut kernel pattern".
The starfield pattern as described by Parizel et al. refers to the presence of scattered bright spots in a dark background in DWI with diffusion restriction, was considered positive if it was identifiable in the presented DWI and ADC images, or was clearly described in text [1]. Cases referred to as starfield positive without presented ADC or clear reference to diffusion restriction, but with very typical starfield pattern in the presented DWI agreed by the three reviewing radiologists in the present study were also accepted as definitive starfield pattern. We categorised a case as starfield negative if there was no mention or image of the starfield pattern, ADC map did not show restricted diffusion, and there was no additional mention of restricted diffusion in the text. Confluent diffusion restriction in the corpus callosum, if present, was also recorded.
Further recorded imaging findings included confluent cytotoxic edema in the white matter (other than corpus callosum), vasogenic edema in the white matter, petechial hemorrhages of non-characteristic pattern, and chronic sequel -atrophy.

Data analysis
Descriptive statistics were used to summarise the recorded parameters. To assess the presence rate of "walnut kernel microbleed pattern" for CFE, the number of characteristic CFE pattern cases was divided by the number of all cases where it was possible to evaluate the presence of microbleed pattern. Similarly, presence rate of starfield pattern for CFE was calculated by dividing the number of cases with definitive starfield pattern by all cases in which it was possible to definitively evaluate. Characteristic microbleed pattern and starfield pattern presence rates were compared using Fisher exact test. For a more direct, one-sample comparison, the presence rates were compared by McNemar test in a subset of patients for whom it was possible to evaluate both the microbleed and the starfield patterns. Both Fisher and McNemar tests were run in MedCalc Statistical Software version 18.11.3 (MedCalc Software bvba https:// www. medca lc. org; 2019, [19]) and tests were regarded significant if yielding a p value less than 0.05.
To present temporal features of the walnut kernel microbleed pattern and the starfield pattern, the included cases were assigned into three stages: acute stage (first 4 days of hospitalisation), subacute stage (5-14 days) and late stage (14 + days). In all stages, four groups were created: starfield pattern positive and negative groups and walnut kernel pattern positive and negative groups. For this analysis, we disqualified case studies and reviews if the exact time point of the MRI for each patient was not reported.

Results
The literature search identified 277 articles, from which 193 articles were excluded, resulting in a total of 84 included articles, for details see Fig. 1 flow chart. The oldest included paper was published in 1998 [15]. In our institution's database search, we have found one case that was tagged by the ICD-10-CM diagnosis code T79.10 for traumatic fat embolization. This was the case of a previously healthy 16-year-old female patient who was run over by a car and suffered severe thoracic and abdominal injuries including right sided serial rib fractures, right sided hemothorax, bilateral pneumothorax, diaphragmatic and hepatic rupture. No direct head injury was recorded based on clinical evaluation and the negative admission CT. Due to delayed awakening on the sixth postoperative day, a head CT was performed which showed multiple punctiform hyperdensities which disappeared by day 17 (Fig. 2.). The typical starfield pattern of CFE, were not seen on the MR images that were acquired on day 12 after admission to the hospital, but there was facilitated diffusion in the subcortical white matter, and diffusion restriction was limited to the corpus callosum only (Fig. 3.). The case showed the walnut kernel microbleed pattern in SWI, surrounded by extracellular edema, except for the corpus callosum where cytotoxic edema was shown (Figs. 3, 4). Considering the results of lab tests and previous reports, the diagnosis of cerebral fat embolism was made. Following supportive care and three months long rehabilitation, a near-complete neuropsychological recovery was achieved. Long term follow-up MR images 16 months later revealed signs of cerebral atrophy (Fig. 5.).
From the 84 included articles, there were 73 case reports, 6 review papers, 4 original papers, and 1 review  pictorial essay. 140 patients from the included articles satisfied our inclusion criteria and were included in this review, in addition to the previously mentioned case from our own databases, resulting in an overall patient number of 141.
The extracted epidemiological and clinical data are summarised in Table 2.
In the included literature SWI was done in 40 cases, T2* was performed in 21 cases, and both SWI and T2* measurements were performed in 12 cases. Microbleeds in general, were present in 52 out of 53 cases (98.11%). The "walnut kernel microbleed pattern" was found in 35 out of 39 evaluable cases resulting in a presence rate of 89.74%.
Diffusion abnormality in general, was seen in 97.64% of the cases (n = 124). The starfield pattern description was used in 70.08% (n = 89) of the cases, but only 19.10% (n = 17) of these cases were confirmed with an ADC map (Fig. 6.). The majority of the presented cases relied on T2 or DWI images only for using the starfield pattern term. However, in 59 out of these cases, diffusion restriction was referred to in the text, in other 8 cases the authors did not use the term starfield pattern, yet it was clearly present in the shown images, and in 3 cases the authors described the lesions as starfield positive, but there was no further accurate detailing of this finding presented. Thus, definitive starfield pattern was ascertained in 87 out of 127 cases yielding a presence rate of 68.5%. In 3.3% (n = 4) the term was used to describe foci of vasogenic edema, large confluent areas of cytotoxic edema, or infarcts were described on T2 images alone [20][21][22][23].
Between the presence rates of the walnut kernel and definitive starfield patterns (89.74% vs. 68.5%), Fisher exact test showed a significant (p = 0.0073) difference. In the subset of patients where both microbleed and starfield patterns were possible to be evaluated (n = 29), the "walnut kernel pattern" was present in 27 cases (93.1%), while the starfield pattern was present in 12 cases (41.38%). The McNemar test showed these rates to be significantly different (p = 0.0003).
Regarding the temporal characteristics of the lesions (Fig. 7.), the starfield pattern was mostly present within the first 4 days after injury where 21 positive and 6 negative cases existed, while only 3 positive cases were reported along with 3 negative cases in the 4-14 days period. In turn, the walnut kernel microbleed pattern had a more consistent presence among

Discussion
Since the clinical diagnosis of CFE is often difficult, imaging could provide significant help. We hypothesized that the increasingly described microbleeds and their particular distribution, the walnut kernel pattern (diffuse uniform punctiform microbleeds located in the subcortical region, the corpus callosum, and in the internal capsule) maybe an important marker for CFE.
Reviewing the literature and our database, it can be noted that although the authors have previously not referred to this actual pattern of microbleeds, the walnut kernel pattern appears to be indeed very common in CFE, as it was present in 89.74% of the cases. Recognizing this specific pattern could be utmost important because microbleeds in other distributions can be associated with other, possibly coinciding pathologies such as diffuse axonal injury (DAI) or sepsis, but they are also common in cardiovascular diseases.
In contrast to the walnut kernel pattern, microbleeds in DAI are seen in groups with varying size and shape (linear, curvilinear, to ovoid), in specific locations [18]. Hypertension associated microbleeds are typically seen only in deeper areas of the brain, while in amyloid Table 3 Radiological findings *Total number of cases eligible for analysis **For definition see methods a Cases with presented susceptibility-or T2* images, or with no presented images but with clear description regarding microbleed presence b Cases with susceptibility-or T2* images in which the subcortical white matter, internal capsule, and the corpus callosum were evaluable, and cases with no presented images but description of findings regarding microbleeds in the specified locations c Cases with presented DWI, or DWI and ADC. Cases with no such presented imaging but with clear description of any diffusion abnormality, or the absence of any diffusion abnormality were also included d Cases with presented DWI and ADC images. Cases without presented images but with description of findings regarding the presence or absence of confluent cytotoxic edema were also included e Cases with DWI and ADC images where the corpus callosum is visible. Cases without presented images but with clear description regarding the presence or absence of corpus callosum diffusion restriction were also included f Cases with presented DWI and ADC images. Cases without presented images but with clear description regarding the presence or absence of lesions of facilitated diffusion were also included. g Cases with presented follow-up MR or CT images. Cases without presented images but with clear description regarding the presence or absence of atrophy in the late stage
Diffuse cerebral microbleeds have been reported in up to 2% of patients receiving extracorporeal membrane oxygenation (ECMO) [33]. While having a similarly favourable prognosis to CFE in the published cases, diffuse cerebral microbleeds related to ECMO also have a quite similar distribution, therefore it is very important for radiologist to be informed about the patient undergoing such therapy [34,35]. Disseminated cerebral microhemorrhages have been described as a late complication in critically ill COVID-19 patients as well [36]. When compared to CFE, microbleeds in critically ill patients such as in sepsis, disseminated intravascular coagulopathy, COVID-19 infection, and patients undergoing ECMO therapy tend to be more variable in size, and shape [32,[36][37][38]. To date, the exact pathomechanism of cerebral microbleeds in the critically-ill in general and in patients undergoing ECMO therapy is not known, but a commonly accepted speculation is that it is probably multifactorial in its nature, involving an ischemic and/ or a biochemical insult induced a thrombotic or embolic event, hence the authors speculate that the underlying pathomechanism could be similar to that of CFE. Further imaging research and accompanying post-mortem studies might elucidate differential diagnosis and the exact mechanisms of microbleeds in critically ill patients.
Beyond microbleeds several types of CFE MR signs have been described in previous studies. According to a comprehensive review published in 2014, the imaging patterns of CFE were classified in to 4 groups (scattered cytotoxic edema, confluent cytotoxic edema in white matter, vasogenic edema lesions that may enhance, petechial hemorrhage of white matter, and chronic sequel), amongst which the starfield appearance was the most commonly encountered (61.5% of 44 MRI scans) [17]. In our review, it was possible to include more recent papers and therefore a larger number of patients were analysed that revealed a very similar occurrence rate 68.5% of the starfield pattern in CFE. However, the more recent papers included also considerably more SWI and T2* studies, revealing that microbleeds, and specifically the walnut kernel pattern may be even more common than the starfield pattern.
An emerging concern regarding the starfield pattern seems to be that there is a degree of uncertainty around its use in the literature. Originally, Parizel et al. has described the starfield pattern as "scattered bright spots on a dark background" on DW-MRI, which, according to their interpretation, "presumably reflect foci of cytotoxic edema" according to low ADC values [1]. Since then this expression has been used widely in the literature, but often without considering the ADC values. According to our findings this was the case in around half of the reviewed patients. In a few published cases the starfield pattern was used to describe confluent T2 hyperintense white matter lesions potentially representing other pathologies like vasogenic edema, while in other reports lesions of obviously high ADC values were referred to as starfield pattern.
From a practical point of view, it is important to note that compared to the starfield pattern, over time, microbleeds could be more consistently identified on MRI. Although due to the small number of corresponding cases, reasonable statistical analysis was not possible, it seems that the starfield pattern is typically present in the first 4 days, as there are much more positive than negative cases in this period, whereas in the 4-14 day time interval the number of positive and negative cases are equal. In contrast, irrespective of the time point, microbleeds were always more commonly present than absent-indicating constant visibility (Fig. 7). This is in line with the article published by Kuo et al. [17].
An additional remarkable observation of our review is that the previously described "confluent cytotoxic edema" is mostly present in the corpus callosum, with a prevalence of 77.27%, which is comparable to the starfield and walnut kernel patterns.
There is limited data in the literature regarding the chronic radiological sequel of CFE. Although the long term prognosis of CFE is generally considered to be good, cerebral atrophy and persistent T2 hyperintensities can be seen in some cases where follow-up imaging has been performed [1,17,[24][25][26][27][28][29].
This study has certain limitations. First, comparison of the rates of the starfield and walnut kernel patterns is somewhat limited since considerably more articles investigated diffusion abnormalities than microbleeds. Second, the presence rates of the investigated imaging patterns do not necessarily reflect their true sensitivity, since it is possible that authors published their imaging results more likely in case of positive findings, causing seemingly higher presence rates of these imaging markers. Still, it can be safely postulated that the walnut kernel microbleed pattern, the starfield pattern and corpus callosum diffusion restriction are the most common imaging alterations in CFE. Third, data regarding chronic changes induced by CFE are scarce and therefore the prevalence of cerebral atrophy remains unknown.

Conclusion
Microbleeds are very common and mostly occur in a characteristic pattern resembling a "walnut kernel" in the CFE MRI literature. Microbleeds of this pattern in SWI or T2* MRI, along with the starfield pattern and corpus callosum diffusion restriction in DWI/ADC appear to be the most important imaging markers of CFE and may aid the differential diagnosis in clinically equivocal cases.