Clinical Radiology (2007) 62, 564e572

Bacterial brain abscesses: prognostic value of an imaging severity index

M.K. Demira,*, T. Hakanb, G. Kilicogluc, N. Cerand, M.Z. Berkmanb, ktasd I. Erdemd, P. Go
Department of aRadiology, Trakya University School of Medicine, Edirne, Turkey, Departments of b Neurosurgery, cRadiology, and dInfectious Disease, Haydarpasa Numune Education and Research Hospital, Istanbul, Turkey
Received 28 June 2006; received in revised form 3 January 2007; accepted 4 January 2007

AIM: To assess the correlation between imaging ndings [computed tomography (CT) or magnetic resonance imaging (MRI)] and neurological status before and after the treatment of bacterial brain abscesses. MATERIALS AND METHODS: CT and MRI images of 96 patients with brain abscesses were retrospectively evaluated in terms of the number, location and size of lesions, and the presence and extent of perilesional oedema and midline shift. An imaging severity index (ISI) based on these different radiological parameters was calculated. Initial Glasgow Coma Scale (GCS) scores and ISI were assessed and the prognostic value of these two indices was calculated. The Pearson correlation test, ManneWhitney test, Chi-square test, receiver-operating characteristic (ROC) analysis, together with comparison of ROC analyses and Fishers exact test were used. RESULTS: There was a negative correlation between ISI and the initial GCS values: ISI increased as the GCS score decreased, indicating an inverse relationship (r 0.51, p < 0.0001). There was a signicant difference between the ISI and GCS scores of patients with an adverse event compared with patients with good recovery. Outcome was signicantly worse in patients with initial ISI over the calculated cut-off values of 8 points or GCS scores under the cut-off value of 13 points. CONCLUSION: ISI is a useful prognostic indicator for bacterial brain abscess patients and correlates strongly with the patient outcome for all parameters studied. ISI score had a better prognostic value than GCS. 2007 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.

Introduction
Brain abscess (BA) is dened as a focal suppurative process within the brain parenchyma.1 In developed countries, BA occurs with an estimated incidence of 0.9 per 105 person-years and with a mortality of up to 20%.2 The clinical manifestations can be life-threatening or cause long-term neurological decits. However, as the symptoms and signs are non-specic, diagnosis and treatment of this potentially treatable condition may be delayed.
* Guarantor and correspondent: M.K. Demir, Trakya University School of Medicine-Edirne-Turkey, Radiology, 11. kisim, Yasemin Apt, D. Blok. Daire 35 Atako y, Istanbul 34158, Turkey. Tel.: 90 533 553 12 46; fax: 90 212 219 62 44. E-mail address: demirkemal@superonline.com (M.K. Demir).

Besides the necessity to make a rapid diagnosis, monitoring the clinical response of a BA is very important in providing appropriate treatment. With the exception of some epidemiological data showing signicant variations, no signicant changes occurred in the prognosis of patients with BA despite improvements in diagnostic procedures and treatment.3 However, information provided by imaging techniques may be used to help the clinician in choosing the correct treatment method and foreseeing its effects. The aim of the present study was to investigate the possibility that imaging may provide an insight into the prognosis of the patient at initial diagnosis by retrospectively analysing the imaging studies of 96 BA patients diagnosed over 13 years.

0009-9260/$ - see front matter 2007 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.crad.2007.01.005

Bacterial brain abscesses

565

Materials and methods

Patients
The study protocol was approved by the institutional review board, which did not require informed consent to be obtained. Ninety-six consecutive patients with clinically diagnosed BA, whose imaging studies and treatment were carried out in our institution, were included in the study. The study group consisted of 66 male and 30 female patients, mean age 29.7 years (range 5e80 years), with 29 patients (30%) being under the age of 15 years. All patients underwent initial and follow-up CT examinations and/or MRI studies. The clinical course of each patient was carefully reviewed with emphasis on presenting clinical symptoms, aetiological and predisposing factors, laboratory ndings, results of neurological examination, and treatment options. The type and duration of antibiotic therapy, together with that of corticosteroids, were also documented. When necessary, surgical methods, e.g., CT-guided aspiration, surgical resection or cerebrospinal uid diversion, were performed and the abscess uid obtained via these methods was examined microbiologically and cultured for aerobic, anaerobic bacteria, mycobacteria and fungi with use of standard culture methods. Patients were clinically followed-up for a median of 9 months (range 3e84 months). Details of clinical ndings and results of the patient group were published elsewhere previously.4

events, no matter what the extent was, as bad outcome and all statistical tests were based on this two-group classication.

CT and MRI
CT examination was the only imaging examination undertaken in 55 patients, while the other 41 patients were examined with CT and MRI. CT examinations of the patients were performed on single-section spiral CT machine (Pronto, Hitachi, Tokyo, Japan and General Electric, Prospeed Plus, Milwaukee, WI, USA). CT sections were obtained as transverse sections at a 15 angle to the orbitomeatal line with a thickness of 5 mm for the posterior fossa and 10 mm for supratentorial regions. Both unenhanced and contrast-enhanced images were obtained. All contrast-enhanced CT examinations were performed after an intravenous injection of iodinated contrast medium (1 ml/kg body weight for children and 50 ml for adults). In 41 patients who were studied with MRI, 26 were imaged at 1.5 T and the remaining 15 at 1 T. MRI studies consisted of spin-echo (SE) T1weighted (T1W), fast SE (FSE) T2W, and postgadolinium SE T1W images (after a dose of 0.2 ml/kg body weight) in the axial, coronal, and sagittal planes with 3 to 6 mm contiguous sections over the whole brain. MRI spectroscopy and diffusion-weighted MRI images were used in the differential diagnosis in several cases.

Radiological scoring
All initial CT and MR images were evaluated retrospectively and the following radiological features were assessed and scored: number, location, and largest diameter of the abscesses, presence of surrounding oedema and presence of midline shift. The scoring system used in the study is summarized in Table 1. When abscesses were single, multilocular or attached to each other like a bunch of grapes, they were scored a single point, the score increased by one point for every additional abscess. Localization was classied as supercial (one point) when the lesion was located in the cerebral and cerebellar hemispheres, deep (two points) when it was in the basal ganglia, thalamus, corpus callosum, brain stem, vermis or within the ventricle, and extensive (three points) when the abscess involved both supercial and deep structures. In the presence of multiple abscesses, the highest score for localization of an abscess was used. The diameter of the abscess was dened as the largest length in the transverse plane on CT images and MRI studies. The largest diameter of

Clinical parameters
Neurological status of the patients was graded at initial presentation according to the Glasgow Coma Scale (GCS)5 and at discharge from the hospital, according to the Glasgow Outcome Scale (GOS).6 GCS is scored between 3 and 15, 3 being the worst and 15 the best. It is composed of three parameters, including best eye response, best verbal response, and best motor response. A coma score of 13 or higher correlates with a mild brain injury, 9 to 12 a moderate injury and 8 severe brain injury. The ve categories of the original outcome scale are: dead, vegetative, severely disabled, moderately disabled, and good recovery. An extended version of the scale, which is not used in this study, divides each of the latter three categories in two, constituting eight categories. To provide simplicity in statistical evaluation, patients without any adverse event were referred to as outcome of good recovery and those with adverse

566

M.K. Demir et al.

Table 1 Calculation of the imaging severity index of patients with bacterial brain abscesses Parameters Number Solitary Multiple Location Supercial Deep Combined Diameter <2 cm 2e4 cm >4 cm Points 1 2e6 1 2 3 1 2 3

analysis. (4) Is the likelihood of a bad outcome signicantly higher in patients with initial ISI or GCS scores under the calculated cut-off values? Chi-square test or Fishers exact test were used to answer this question. (5) Which score has the better prognostic valuedISI or GCS? Comparison of ROC curves method was used to compare prognostic values of ISI and GCS. Statistical tests were performed using MedCalc Statistical software (version 7.2.0.2, Mariakerke, Belgium). The signicance level was set at p < 0.05.

Results
The detailed presenting clinical signs and symptoms, aetiological and predisposing factors, laboratory data, isolated micro-organisms, treatment methods, and statistical analyses of correlation between these variables were published previously.4 The major treatment preferences for the patients are listed in Table 2.

Perilesional oedema Minimal (maximum thickness < radius of abscess) 1 Moderate (maximum thickness between the 2 radius and diameter) Large (maximum thickness > diameter of abscess) 3 Midline shift Mild (<5 mm) Moderate (5e10 mm) Severe (>10 mm) 1 2 3

Clinical results
a multiloculated abscess was measured by taking the largest diameter of the lesion taken as a whole. Surrounding oedema, which is observed as high signal intensity on T2W MRI images or hypo-attenuating areas on CT images, was graded as minimal, moderate or large if its maximum thickness was less than the radius of abscess, between the radius and diameter and more than the diameter, respectively. In the presence of multiple abscesses, the highest score for oedema was taken into account. Midline shift was classied as mild (one point), moderate (2 points) or severe (3 points) if it was <5 mm, between 5 and 10 mm and >10 mm, respectively. For each patient a total imaging score [imaging severity index (ISI)] was calculated by adding scores obtained from each parameter. GOS results showed that eight patients (8%) died despite treatment due to intraventricular rupture of the abscess, late adverse events of traumatic abscess formation, pulmonary failure or cardiac insufciency. Ten patients had severe disability, among whom one patient had unilateral visual loss, two had dysphasia and seven had hemiparesis or hemiplegia. Twenty-three patients had moderate disability and 55 patients recovered completely (Table 3).

Imaging ndings
One hundred and twenty-three lesions were observed using CT and MRI. In 80 (83%) patients, lesions were solitary, while multiple abscesses (two to six abcesses) were found in 16 patients (17%). Solitary abscesses were mostly located in the frontal and temporal lobes. Abscesses originating from sinusitis and dental infection were usually located in the frontal lobe, and in those

Statistical analyses
Statistical analyses were used to answer ve questions: (1) is the ISI value correlated to the initial GCS result? The Pearson correlation test was used for this evaluation. (2) Is there a signicant difference between the ISI or GCS scores of patients with an adverse event compared with patients with good recovery? ManneWhitney test was used for this comparison. (3) What are the cutoff values for ISI or GCS scores, to predict a bad outcome? These values were calculated using receiver-operating characteristic (ROC) curve

Table 2 Treatment methods of patients with bacterial brain abscesses Treatment method Surgical Aspiration Excision Non-surgical Number (%) 86 72 14 10 (90%) (84%) (16) (10%)

Bacterial brain abscesses

567

Table 3 Use of Glasgow Coma Scale (GCS) and outcome of patients after treatment in the study, with comparison of initial GCS score and Glasgow Outcome Score Initial scores according to GCS Scores No. of patients 28 18 25 8 5 3 1 4 2 Outcome of patients after treatment Recovery Moderate Severe Death adverse adverse event event 19 12 17 4 2 7 5 5 3 1 1 2 1 2 1 2 2 1 1 1 1 1 2

Scores obtained from the analysed parameters, initial GCS scores at diagnosis and outcome of each patient are listed in Table 4 and representative examples of the cases are shown in Figs. 1e5.

Statistical analyses
Is the ISI value correlated to the initial GCS result? Pearson correlation test revealed a negative correlation between the radiological score (ISI score) and initial neurological assessment, indicating an inverse relationship (r 0.51, p < 0.0001; 95% condence interval (CI) for r 0.64 to 0.34). Is there a signicant difference between the ISI or GCS scores of patients with an adverse event compared with patients with good recovery? ISI scores of patients with no adverse event (median 7; 95% CI for the median 7e8) were signicantly lower compared with patients with an adverse event (median 11; 95% CI for the median 9e12; p < 0.0001). Similarly, a signicant difference (p 0.005) was present between GCS scores of patients with an adverse event (median 13; 95% CI for the median 11.9e14.0) or good recovery (median 14; 95% CI for the median 13e14.1). What are the cut-off values for ISI or GCS scores, to predict a worse outcome with an adverse event? ROC curve analysis [area under the curve (AUC) 0.884 0.004] calculated the cut-off value for the ISI score as 8 points, inferring that cases with an ISI score of 9 or higher have a higher probability for any kind of adverse event (sensitivity 92.7%, specicity 69.1%). The cut-off for the GCS score was calculated as 13 points (sensitivity 43.9%, specicity 87.3%; AUC 0.670 0.055), that is, patients with a GCS score of 12 or

15 14 13 12 11 10 9 8 7 6 5 4 3

2 96 55 23

1 10

1 8

that were related to otitis and mastoiditis, location was mostly cerebellar and temporal. Of the 123 abscesses, 82 were located supercially (67%), 14 deep (11%) and the other 27 extensively (22%). Diameter of abscesses ranged from 8e67 mm (mean 32.8 mm) being >2 cm in 40 (32.5%), between 2e4 cm in 56 (45.5%), and >4 cm in 27 (22%) of the 123 brain BAs. In each case, brain oedema was detected surrounding the abscess. Perilesional oedema was rated as minimal in 27 patients (28%), moderate in 35 (37%), and large in 34 (34%). In terms of midline shift, there were 39 patients (41%) in the rst group with no or minimal shifting, 24 patients (25%) in the moderate group, and 33 patients (34%) in the large group.
Table 4

Comparison of imaging severity index scores and outcome of patients after treatment in the study Outcome of patients after treatment Recovery Moderate adverse event 1 3 2 4 1 10 2 1 1 2 2 2 2 1 3 1 1 1 Severe adverse event Death 1

Initial imaging severity index scores Scores 16 15 14 13 12 11 10 9 8 7 6 5 No. of patients 1 1 2 7 7 8 10 19 10 15 10 6 96

2 1 6 7 8 15 10 6 55

23

10

568

M.K. Demir et al.

Figure 1 Supercial bacterial BA with ISI 6. The patients GCS score was 14. (a) Coronal TSE T2W [repetition time/ echo time (TR/TE) 4461 ms/100 ms], and (b) gadolinium-enhanced transverse T1W SE (TR/TE 360 ms/6 ms) MRI images show a mass (arrows) with a hyperintense centre ()) and surrounding moderate oedema (dot in a and b) in the left temporal lobe. (c) Contrast medium-enhanced transverse CT image during medical treatment reveals regression in left temporal lobe abscess (arrowhead) and resolution of surrounding oedema. The patient recovered without any adverse event.

less have a higher probability for any kind of adverse event. Is the rate of a worse outcome signicantly higher in patients with initial ISI over or GCS scores under the calculated cut-off values? Only three out of 41 patients with an ISI score of 8 demonstrated any adverse event, whereas 38 of 55 patients having ISI scores of !9 showed some disability. Fishers exact test proved this difference to be highly signicant (p < 0.00001). Adverse events were present in 23 of 71 patients having GCS scores of 13 or higher, whereas 18 out of 25 patients with GCS scores of 12 had adverse events. Chi-square test demonstrated that this difference was also statistically signicant (p 0.0013). Moreover, these results show that using the cut-off values indicated in the study, groups demonstrate signicant differences. Which score has a better prognostic value? ISI or GCS? When initial GCS and ISI scores were compared as predictors of outcome using the ROC curves method, ISI was a signicantly more consistent method than GCS (p < 0.0001; difference between areas 0.214 0.051; 95% CI: 0.115e0.313). This means that in predicting the nal outcome for the patient after proper management, ISI was a better indicator.

Discussion
An intracranial abscess is dened as pyogenic debris accumulating in the brain and is a life-threatening

medical emergency. Successful treatment of the disease depends on early diagnosis, timely treatment, and vigilant monitoring of response to treatment. After BA is diagnosed, the decision for surgical versus medical treatment must be made; decisions are usually made using a combination of the imaging, clinical and laboratory ndings. Until now, no unique prognostic indicator for BA has been reported. This is the rst investigation to produce a prognostic indicator for bacterial BAs, which uses the elementary imaging ndings as a scoring system (ISI). It is well known from previous studies that imaging ndings such as size, location, number (multiple versus single), and type (multiloculated versus uniloculated) may inuence the treatment of choice for bacterial BAs. Although some authors mention that the size of a BA is an important factor in planning initial therapy,7,8 our colleagues did not nd signicant correlation between the size and prognosis.4 Large abscesses may be refractory to antibiotic treatment alone. Some authors selected medical treatment for abscesses smaller than 2 cm,9 2.5 cm,10 or between 2e3 cm.11 Prognosis of the disease may also be worse in multiple abscesses, in intraventricular rupture of abscesses, and in those that are deeply located. The present study showed that the ISI scoring system improves the early prognostic value of CT and MRI in cases of bacterial BAs. Moreover, statistical evaluations revealed that it is much more consistent than the GCS. As the ISI scores increase, the severity of the abscesses escalates

Bacterial brain abscesses

569

Figure 2 Supercial bacterial BA with ISI 8. (a) Contrast medium-enhanced transverse CT image shows a ring-enhancing mass (arrow) with a hypoattenuating centre () ) and surrounding mild oedema (dot) in the left temporal lobe. (b) Contrast medium-enhanced transverse CT image after treatment reveals regression in abscess mass (arrow), although surrounding oedema and midline shift not only persisted, but increased in severity. The patient recovered with moderate adverse event.

Figure 3 Supercial bacterial BA with ISI 10. The patients GSC score was 15. (a) Contrast medium-enhanced transverse CT image shows a ring-enhancing mass (arrow) with a hypoattenuating centre ()) and surrounding mild oedema (dot) in the right temporal lobe. The abscess causes severe midline shift (arrowhead). (b) Contrast medium-enhanced transverse CT image after treatment reveals a hypoattenuating area (thick white arrow). The patient recovered with severe adverse event.

570

M.K. Demir et al.

Figure 4 Multiple bacterial BAs with ISI 13. The patients GSC score was 14. (a) Contrast medium-enhanced transverse CT image through the level of posterior fossa shows ring-enhancing abscesses in the left cerebellar hemisphere and left temporal lobe (arrows). (b) Contrast medium-enhanced transverse CT image through the level of third ventricle shows multiple abscesses in the left temporal and parietal region adjacent to each other (arrows). According to the ISI system this is accepted as a single abscess and the patient is assigned two points from the number of abscesses, three points from the size as the largest abscess diameter is over 4 cm, three points from the location as the abscess extends to left nucleus lentiformis, two points from the surrounding oedema as its width is not more than the abscess diameter (although it is less than the radius of abscess in this section, the surrounding oedema is larger in supraventricular levels), and three points from the midline shift as it is >1 cm, making the ISI score 13. (c) Contrast mediumenhanced transverse CT image during treatment reveals regression of the abscess size (arrow), surrounding oedema (dot) and the other imaging parameters. The patient recovered without any adverse event.

with a corresponding increase in neurological decits and mortality. ISI is straightforward and reproducible because only ve imaging parameters are used: number, location, size, amount of surrounding oedema, and extent of midline shift. The major advantage of the presented ISI is that it can be used by both clinicians and radiologists. Assessment is very simple and there is no need to make detailed measurements. It would also be interesting to determine which among all the ve items constituting ISI was the most relevant in terms of prognostic value. However, any effort to determine the leading indicator among those individual radiological factors failed, and none of the items was found to be more important or dominant relative to each other for the patients prognosis. There are many factors inuencing treatment decisions for a strategy that would result in the best possible clinical outcome for a BA patient. Initial neurological status is the most important factor on prognosis and mortality, which has been evaluated with GCS in various settings.7 Prognosis is signicantly poor for patients presenting with lower GCS scores. In the present study 12 (86%) of the 14 patients with initial GCS score <10 either died or became vegetative. This is consistent with the results from other studies.12,13 Conversely, only

two (4%) of the 46 patients with initial GCS scores of 14e15 became vegetative. There were no deaths in this group. For most patients, the pretreatment neurological state is an important prognostic indicator related to the outcome. This nding is conrmed in the present study, where GCS was found to be signicantly correlated with the nal outcome and a cut-off value for GCS was dened. There were some limitations in the present study. The ISI system depends solely on imaging characteristics and evaluates only the BA. It is not inuenced by the severity of underlying cause of the BA. Some of the bad outcomes listed in the study were actually due to other diseases outside of the brain. Such various external parameters cause an inevitable lack of control when judging the outcome. Further investigations may be needed, i.e., control groups that create cohorts for similar pathogens, antibiotic sensitivity versus resistance, or similar treatment regimens, such as surgery versus no surgery. The nature of the causative organism and the patients immunocompetency are important factors in predicting outcome. However, in the present study, there were no immunocompromised patients or any causative agents such as Aspergillus spp., which might have a worse outcome even with a low initial ISI. Another potential problem is the length of time of

Bacterial brain abscesses

571

Figure 5 Multiple bacterial BAs with ISI 16. Gadolinium-enhanced transverse T1W SE (TR/TE 360/6) MRI images (a) through the level of posterior fossa and (b) through the level of lateral ventricles show multiple ring-enhancing masses (arrows) with a hypointense centre ()). The masses are either located supercially and deep. (c) Contrast mediumenhanced transverse CT image shows two supercial (arrows) and one intraventricular abscess (arrowhead) with a hypoattenuating centre ()) and surrounding mild oedema (dot). The abscesses capsules are slightly enhancing (curved arrow). There is severe midline shift. (d) Contrast medium-enhanced transverse CT image during the treatment shows regression in both supercial abscesses masses (arrows), but progression in intraventricular abscess mass ()). Gas is seen inside the abscess cavity due to aspiration (arrowhead). The patient died during the treatment.

572

M.K. Demir et al.

the study period, as the disease processes and treatment options may have changed in that time and may have effected the outcome. In conclusion, the ISI is a useful prognostic indicator for bacterial BA patients and correlates strongly with outcome for all parameters studied. Using such a simple and straightforward method to predict outcome, preferably together with GCS, may provide a common language among professionals, and shorten and minimise any delay in instituting treatment.

References
1. Habib AA, Mozaffar T. Brain abscess. Arch Neurol 2001;58: 1302e4. 2. Calfee DP, Wispelwey B. Brain abscess. Semin Neurol 2000; 20:353e60. 3. Tonon E, Scotton PG, Gallucci M, et al. Brain abscess: clinical aspects of 100 patients. Int J Infect Dis 2006;10: 103e9. 4. Hakan T, Ceran N, Erdem I, et al. Bacterial brain abscesses: an evaluation of 96 cases. J Infect 2006;52:359e66.

5. Teasdale G, Jennett B. Assessment of coma and impaired consciousness. A practical scale. Lancet 1974;2:81e4. 6. Jennett B, Bond M. Assessment of outcome after severe brain damage. Lancet 1975;1:480e4. 7. Tseng JH, Tseng MY. Brain abscess in 142 patients: factors inuencing outcome and mortality. Surg Neurol 2006;65: 557e62. 8. Yamamoto M, Fukushima T, Hirakawa K, et al. Treatment of bacterial brain abscess by repeated aspirationdfollow up by serial computed tomography. Neurol Med Chir 2000;40: 98e104. 9. Mamelak AN, Mampalam TJ, Obana WG, et al. Improved management of multiple brain abscesses: a combined surgical and medical approach. Neurosurgery 1995;36: 76e86. 10. Rosenblum ML, Mampalam TJ, Pons VG. Controversies in the management of brain abscesses. Clin Neurosurg 1986;33: 603e32. 11. Schielke E. Bacterial brain abscess. Nervenarzt 1995;66: 745e53. 12. Xiao F, Tseng MY, Teng LJ, et al. Brain abscess: clinical experience and analysis of prognostic factors. Surg Neurol 2005;63:442e50. 13. Takeshita M, Kagawa M, Izawa M, et al. Current treatment strategies and factors inuencing outcome in patients with bacterial brain abscess. Acta Neurochir 1998;40: 1263e70.