European Journal of Radiology

Transcription

1 European Journal of Radiology 76 (2010) Contents lists available at ScienceDirect European Journal of Radiology journal homepage: Automatic bone and plaque removal using dual energy CT for head and neck angiography: Feasibility and initial performance evaluation C. Thomas a,,a.korn b,b.krauss c, D. Ketelsen a, I. Tsiflikas a, A. Reimann a, H. Brodoefel a, C.D. Claussen a, A.F. Kopp a, U. Ernemann b, M. Heuschmid a a Department of Diagnostic and Interventional Radiology, University of Tübingen, Hoppe-Seyler-Straße 3, Tübingen, Germany b Department of Neuroradiology, University of Tübingen, Tübingen, Germany c Siemens Medical Solutions, Forchheim, Germany article info abstract Article history: Received 23 January 2009 Received in revised form 1 April 2009 Accepted 4 May 2009 Keywords: Dual energy CT CTA Carotid Head and neck angiography Bone removal Purpose: We sought to evaluate the feasibility and efficiency of dual energy (DE) bone and plaque removal in head and neck CT angiography. Materials and methods: 20 patients with suspected carotid stenoses received head and neck DE-CTA as part of their pre-interventional workup. Visual grading using multiplanar reformations (MPR), thick slab maximum intensity projections (MIP) and quantitative vessel analysis (QVA) of stenoses was performed prior and after DE bone removal. Results were evaluated for the detection of relevant stenoses (vessel area reduction >70%). Vessel segmentation errors were analyzed. Results: Segmentation errors occurred in 19% of all vessel segments. Nevertheless, most post-bone removal artifacts could be recognized using the MPR technique for reading. Compared to MPR reading prior to bone removal, sensitivity, specificity, positive and negative predictive values after bone removal were 100%, 98%, 88% and 100% for MPR reading and 100%, 91%, 63% and 100% for exclusive MIP reading, respectively. There was a good agreement between the QVA results prior and post-de plaque removal (r 2 = ). Conclusion: DE bone and plaque removal for head and neck angiography is feasible and offers a rapid and highly sensitive overview over vascular head and neck studies. Due to a slightly limited specificity of the MIP technique due to segmentation errors, possible stenoses should be verified and graded using MPR techniques Elsevier Ireland Ltd. All rights reserved. 1. Introduction CT angiography (CTA) is an established modality for the evaluation of the head and neck arteries [1 3]. Its advantages in comparison to magnet resonance angiography (MRA) include a high spatial resolution and its fast, uncomplicated and robust nature. Reliable quantification of stenoses is possible also in cases of severe vessel kinking, and advanced visualization tools like multiplanar reformations (MPRs), maximum intensity projections (MIPs) and volume rendering techniques (VRTs) allow advanced reading and presentation of results [4]. Disadvantages besides the issue of exposure to ionizing radiation and iodinated contrast media include its limited ability of differentiation between calcium and contrast media, making a precise and automated quantification of calcified stenoses difficult [5]. Furthermore, in order to perform advanced visualization, osseous structures have Corresponding author. Tel.: address: Christoph.thomas@med.uni-tuebingen.de (C. Thomas). to be removed semi-automatically, which is a time consuming and often unsuccessful process, especially in the skull base and the cervical spine where vessels are situated in close proximity to bones. Alternatively, bone subtraction can be performed, which requires an additional unenhanced examination and thus goes along with an increased dose and misregistration problems [5 7]. The possibility to reliably differentiate calcium from contrast media from the data from only a single examination would therefore have the potential to facilitate the acquisition, postprocessing, reading and presentation of head and neck CTA studies and to save the radiation dosage which is needed for bone subtraction. With a recently introduced dual source CT system it is possible to perform simultaneous CT acquisitions with two different X-ray spectra (dual energy DECT [DECT]) [8,9]. Due to different atomic numbers of iodine and calcium and thus different X-ray attenuation profiles at different energy levels, this technique has the potential to differentiate between both materials. Meanwhile, dedicated postprocessing software has been made available by the manufacturer, allowing automated bone and plaque removal in the head and neck vessels based on the DE attenuation information X/$ see front matter 2009 Elsevier Ireland Ltd. All rights reserved. doi: /j.ejrad

2 62 C. Thomas et al. / European Journal of Radiology 76 (2010) The aim of this study was a first assessment of the feasibility and the efficiency of automated bone and plaque removal for head and neck CTA studies using DECT. 2. Materials and methods 2.1. Patients After agreement of the institutional ethics review board, 20 patients with known or suspected carotid artery disease who were scheduled for stent-protected angioplasty or carotid endarterectomy were included in the study. Written informed consent was obtained from all patients. All patients received a standardized DE- CTA of the head and neck arteries as part of their pre-interventional workup Data acquisition and reconstruction All examinations were performed with a dual source CT system (Somatom Definition, Siemens Healthcare, Forchheim, Germany). The examination region was planned from the aortic arch to the cranial end of the skull. After intravenous injection of 60 ml of iodinated contrast medium (Imeron 400 [iomeprol], 400 mg I/ml, Bracco Altana Pharma, Konstanz, Germany) with a flow setting of 5 ml/s and a saline chaser (50 ml, flow 5 ml/s), bolus tracking was performed in the aortic arch. The DE examination was performed in a caudocranial direction, using default acquisition parameters for carotid CTA (tube A: 140 kv, 80 mas ref, tube B: 80 kv, 234 mas ref, automated tube current modulation [CareDose 4D], collimation 32 mm 0.6 mm for each detector with flying focal spots, rotation time 0.33 s, pitch 0.65). After examination, axial sections (1.0 mm, increment 0.6 mm) were reconstructed using a dedicated non-edge-enhancing reconstruction kernel (D30). Two individual stacks of images for each detector (80 and 140 kv images) and DE mixed images were reconstructed; the latter containing weighted information from both detectors with a weighting factor of 0.4 (40% from the 80 kv images and 60% from the 140 kv images) and thus approximating regular 120 kv images Postprocessing and image analysis Fig. 1 shows a flow-chart of the postprocessing routine used for this study. Fig. 1. Flow-chart summarizing the setup of the study. Results of MPR eyeballing (visual subjective approach) and MIP eyeballing post-bone and plaque removal are compared to the results of MPR eyeballing prior to bone removal (mixed images); results of quantitative vessel analysis (QVA) post-bone and plaque removal are compared to QVA prior to plaque removal. All images were anonymized and transferred to an offline workstation, which was equipped with a dedicated DE analysis software package (Syngo CT Workplace, Version syngo CT 2008G, Siemens Healthcare) and advanced semi-automatic vessel analysis software (InSpace, Siemens Healthcare). Each patient s cervical vessels were bilaterally partitioned into the following five segments: (1) aortic arch, brachiocervical trunk and origins of the great neck vessels; (2) common carotid artery (ACC); (3) bulbus and extracranial internal carotid artery (ACI); (4) intracranial ACI; (5) vertebral artery. Segments 1 4 were evaluated for possible stenoses (8 segments per patient, 160 segments in the whole study). A scoring system was used for grading (0: no stenosis; 1: <70%; 2: 70 90%, 3: 90 99%, 4: 100%). At first, freely adjustable MPRs of the mixed images were used to assess the stenoses by eyeballing (completely visual and subjective approach, two experienced, blinded readers [C.T. and A.K.] in consensus). No measurement tools were used for eyeballing. The results were used as a standard of reference for the post-bone and plaque removal evaluation. Afterwards, stenoses were additionally quantified using the semi-automatic quantitative vessel analysis (QVA) tool of InSpace by the same two readers. The vessel segmentation thresholds of QVA were adjusted individually to fit the attenuation of the vessels, calcified plaque was excluded manually. Measurements were performed by referencing the maximum axial area of stenosis to the closest further distal non-stenosed axial area of the vessel. Then, automatic DE bone and plaque removal was performed on the datasets. The algorithm works by analyzing the attenuation of each voxel at 80 and 140 kv. Since calcium and iodine have different atomic numbers (z), their attenuation characteristics vary at different energy levels, and both elements can thus be distinguished by analyzing the ratio in HU between 80 and 140 kv acquisitions [10]. The algorithm furthermore uses vessel tracking algorithms in regions where severe image artifacts are detected, e.g. the shoulder region, as those artifacts do not allow a reliable DE analysis. Further filters like fragment removal and blooming reduction are applied automatically and can be disabled by the user. As an output, the software generates a stack of images where all calcium-containing voxels are assigned a value of 1024 HU and thus appear black. Plaques can be switched on and off. The images can be stored as a new series for further postprocessing or can be used to create MIPs or VRTs directly in the DE application. For the study, the postprocessing time was recorded. No adjustments were made to the algorithm as it is supposed to function independently of the user. After postprocessing, the vessel segments described above (10 segments per patient, 200 segments in the study) were examined for segmentation errors, applying a grading system (0: correct segmentation; 1: partial cancelation of vessel in at least one axial slice; 2: full cancelation of the vessel in at least one axial slice). In an additional reading session, stenoses were quantified using exclusively post-bone and plaque removal thick slab (slice thickness 20 cm) DE MIP images. The users were allowed to freely rotate, scale and window the MIPs. Furthermore, MPR eyeballing was performed with the processed images after bone and plaque removal, but with a latency of at least 7 days after the first reading to reduce recall bias. Stenosis quantification with QVA as described above was performed after plaque and bone removal as well. The results of the post-bone removal MPR eyeballing and of postbone removal thick slab MIP reading were compared to the results of MPR eyeballing prior to bone removal, using the latter as gold standard. Sensitivity, specificity, positive and negative predictive values were calculated for the detection of relevant lesions (stenosis >70%), with MPR eyeballing prior to bone removal as gold standard. The Pearson s correlation coefficient was calculated for the QVA results prior and after DE plaque removal.

3 C. Thomas et al. / European Journal of Radiology 76 (2010) the vertebral arteries were found in asymmetrically lean vessels. Beam hardening artifacts of dental metal reduced DE segmentation performance (Fig. 4). In two patients, stents in the internal and common carotid arteries were detected; both stents were patent. The stent lumina were recognized as non-iodine and were accurately removed from the dataset (Fig. 5). However, in one case the small metal markers on the cranial end of one of the stents caused significant beam hardening artifacts, resulting in an underestimation of the lumen. However, as these stents so not represent natural anatomy they were excluded from further evaluation Plaque removal and stenosis quantification Using the MPR eyeballing method on the mixed images before DE analysis, 22 stenoses were classified as relevant (stenosis >70%). After DE bone and plaque removal, 25 stenoses were rated as relevant with MPR eyeballing and 35 stenoses with the MIP technique, as stated in Table lesions were assessed with QVA. 15 of these stenoses were rated >70% before and also after plaque removal. Since no relevant stenoses were missed after DE analysis, high sensitivities compared to MRP reading prior to bone and plaque removal were achieved for both methods. Specificity and PPV were limited with exclusive MIP reading (Table 3). When comparing QVA results before and after DE plaque removal, high levels of agreement were found (r 2 = , Fig. 6). Fig. 2. Thick slab MIP of the head and neck vessels in a 62-year-old patient after DE bone removal. The postprocessing ran fully automated and did not require user interaction. Note the almost complete removal of osseous structures of the scull base. Due to dental artifacts, segmentation errors occurred in the level of the artifacts. Postprocessing time was assessed for pre-bone removal eyeballing and post-bone removal MIP reading. Additionally, the time needed to create (Syngo 3D, Siemens) and read standardized regular MIP images of relevant cervical and intracranial vessels (coronar and axial cerebral MIPs, coronal and parasagital cervical MIPs) as established in our institution was measured. 3. Results The examinations were performed successfully in all patients. All included carotid arteries could be evaluated. No complications or adverse reactions occurred. Bone and plaque removal worked fully automated and did not require any user interaction (Fig. 2). An example of the effect of plaque removal is shown in Fig Bone removal and vessel segmentation Osseous structures of the skull and the cervical vertebral spine were reliably removed from the images in all cases. The effectiveness of the algorithm was principally limited in regions of bone-to-air transitions due to partial volume effects. The large paranasal sinuses were successfully detected by the software and excluded; however the mastoid cellulae with their many small airspaces were not reliably removed in all cases. As well, as the software is designed for head and neck examinations, in some cases the ribs and other thoracic osseous structures partly remained in the datasets. The efficiency of the DE vessel segmentation was assessed in 200 vessel segments. Segmentation errors were detected in 38 segments (19%), 13 deletions were partial deletions and 25 were full deletions (Table 1). The common carotid artery was segmented correctly in 39 of 40 cases. The ten deletions of the origins of the great vessels occurred in the shoulder regions where the vessels were mainly segmented using the vessel tracing algorithm instead of DE material decomposition. Most of the segmentation errors in 3.3. Postprocessing and reading time The average total time required for automatic DE plaque and bone removal was 89 s, consisting of 19 s for loading of the images and 70 s for DE analysis. The average time needed for manual creation of standardized MIP images was 660 s. Mean reading times were 45 s with standardized MIP images, 40 s with post-bone removal MIP images and 230 s for MPR eyeballing prior to bone removal Radiation dose According to the automatic dose calculations performed automatically by the system, an average volume computed tomography dose index (CTDI vol ) of 7.3 mgy and an average dose length product (DLP) of 277 mgy cm were applied for the DE acquisition. 4. Discussion While CTA of the head and neck vessels is an accepted noninvasive modality to assess hemodynamically relevant stenoses, it is limited by the depiction of bones and plaques, which hampers advanced visualization that would be comparable to invasive digital subtraction angiography (DSA) or MRA. To overcome this limitation, several bone removal algorithms have been designed and evaluated in the past, typically depending on an unenhanced examination prior to CTA or a venous phase examination after the arterial phase [5,7,11 14]. Although this technique of bone subtraction CTA yields acceptable results in study settings, two principal limitations go along with this technique: first of all patient movements (head movements as well as soft tissue movements, e.g. by swallowing) in between the acquisitions might lead to registration problems. Nonrigid registration algorithms have been designed to overcome this problem. It has been shown that they have the potential to lead to improved image quality, but cannot eliminate motion artifacts completely [5,13]. Second, the unenhanced acquisition causes additional exposure to ionizing radiation. Although low-dose techniques are preferred, the effective exposure per examination is typically increased by 20 25% [12]. As bone removal

4 64 C. Thomas et al. / European Journal of Radiology 76 (2010) Fig. 3. High-grade stenosis of left proximal internal carotid artery. Curved MPR prior to bone removal (upper left) show a large mixed plaque. In the corresponding thick slab MIP (lower left), the stenoses are obscured by the calcified parts. After DE plaque removal, the calcifications are removed from the datasets (upper right, curved MRP). In the resulting thick slab MIP (lower right), the stenosis can be assessed. is not necessary for the reading, but rather contributes to a better visualization, it is disputable whether the additional exposure is justified. Both of these limitations are principally overcome with the use of DECT, where 80 and 140 kv data are acquired almost simultaneously. In fact, since the X-ray tubes and detectors are mounted at an angle of 90, there is a temporal shift of about 83 ms at a rotation time of 330 ms. We do not expect this delay to be significant in clinical practice. Also, no or only little additional exposure is required, since data from both detectors can be combined to mixed images with an image quality comparable to a Table 1 Segment-based evaluation of DE vessel segmentation (100 segments). Vessel segment Partial deletion Full deletion Total Percentage of vessels [%] Sum Segment 1: Aortic arch/brachiocervical trunk/origins of the great neck vessels; 2: common carotid artery (ACC); 3: bulbus/extracranial internal carotid artery (ACI); 4: intracranial ACI; 5. vertebral artery.

5 C. Thomas et al. / European Journal of Radiology 76 (2010) Table 2 Results of segment-based evaluation of stenosis after DE bone and plaque removal. Method Result Segment Sum MPR rp rn fp fn MIP rp rn fp fn Eyeballing prior to DE plaque removal serves as gold standard. Segment 1: Aortic arch/brachiocervical trunk/origins of the great neck vessels; 2: common carotid artery (ACC); 3: bulbus/extracranial internal carotid artery (ACI); 4: intracranial ACI; rp/fn: right/false positive; rn/fn: right/false negative. Table 3 Sensitivity and specificity after DE bone and plaque removed for all three methods as calculated from Table 2. MPR MIP Sensitivity Specificity PPV NPV Fig. 4. DE vessel segmentation errors due to metal artifacts caused by dental implants: partial deletion of right internal and external carotid artery. Since several vessels are deleted in the same position in the z-axis, the artifact can easily be recognized as such. single energy 120 kv examination at about the same dose level [8]. In this study we report our initial results regarding the potential and the limitations of DE bone and plaque removal in the cervical arteries. The efficiency of vessel segmentation clearly depended on the anatomic region: the best results were achieved in the common and the extracranial internal carotid artery; in these cases the segmentation was only disturbed by metal artifacts due to dental or other metal objects. More deletions occurred in the intracranial parts of the internal carotid artery and in the vertebral artery, especially in very thin vessels in close contact to bones. It has to be acknowledged that very small vessels only occupy few pixels per image and have a low attenuation. When taking image noise into account, an extraction of DE information is quite difficult in Fig. 5. Plaque removal applied to a vessel with stent: accurate removal of the stent lumen (left). The metal markers at the apical end of the stent cause underestimation of the lumen due to beam hardening artifacts (right). Center images: curved MPR, left/right: orthogonal planes to the centerline (the arrows indicate the location); upper row: mixed source images, lower row: images after DE bone and plaque removal.

6 66 C. Thomas et al. / European Journal of Radiology 76 (2010) This step appears inevitable as long as segmentation errors and false positive results occur. In future software versions these problems might be solved which would lead to an improved specificity and thus to a rapidly decreased need for time consuming standardized MIP reconstructions. However, larger studies are needed before such a workflow model can be implemented into clinical routine. This initial technical feasibility study is limited by the absence of a clearly defined gold standard like DSA, which would have helped to better quantify the results. However, the aim of the study was not an evaluation of CTA, but of DE bone and plaque removal on CTA data. Therefore, sensitivity and specificity are given in comparison to MPR reading prior to bone and plaque removal, which nowadays is the standard way of reading CTA studies. Further larger studies with correlation to DSA and MRA are warranted to fully assess the comparability of the results. 5. Conclusions Fig. 6. Correlation of QVA results before and after DE bone and plaque removal. these vessels. In most of the cases, these errors could be visually detected and identified when MPRs were analyzed. However, when MIP visualizations were used exclusively, it was not possible to reliably identify segmentation errors, which caused 13 false positive results in the 20 patients and thus limited specificity (91%) and positive predictive value (63%). This leads to the conclusion that MIPs should not be used exclusively for diagnosis, which corresponds to current diagnostic standards. The accuracy of plaque removal was quantified semi-automatically using a QVA tool. The results show a good agreement of the pre- and post-plaque removal measurements (r 2 of ). Since the reproducibility of QVA is known to be limited and observer-dependent due to the need to define thresholds, this correlation has to be regarded as very high [15 17]. The extent of plaque removal in the case of calcified plaques therefore appears to be calibrated correctly. These observations are also in good agreement with a previous study by Uotani et al. who found a good agreement of post-plaque removal MIP reading results and invasive DSA [18]. Also two stents were removed from the datasets within the known limits of CT [19 21], although beam hardening artifacts of metal markers caused an underestimation of the lumen in one case. DE bone removal ran fully automatic and independently in all cases, no user interaction was required after the bone removal process was started. The mean processing time per patient was around 90 s, which is lower than with other techniques previously introduced: for nonrigid registration for subtraction CT angiography using an unenhanced acquisition, procession times are reported to be around 30 min [5]. Vega Higuera et al. and Abrahams et al. both designed algorithms to remove the skull base from contrast enhanced single energy examinations; however the latter requiring runtimes of min [22,23]. Obviously, different and probably less capable computer platforms were used in these studies, also influencing processing speed. We also assessed the operator time needed to create standardized MIP images, which was relevantly more time consuming (around 660 s). Since the durations of post-bone removal MIP reading and reading of standardized MIPs were comparable and since a high sensitivity for the detection of stenoses with post-bone removal MIP reading was found, post-bone removal MIP reading might be performed in a clinical setting to rapidly rule out stenoses almost instantly after the examination. Time consuming creation of standardized MIPs would be a second step if a possible stenosis is found in the MIPs and might be spared in the case of a previous exclusion of a relevant stenosis. In conclusion, DE bone and plaque removal of head and neck arteries is feasible and yielded promising and highly sensitive initial results, although the technique appeared to be susceptible to artifacts in its current state of development, leading to a reduced specificity and positive predicted value. Examiners should be aware of these limitations and use MPR techniques for the verification and grading of possible stenoses. The application of DE can be expected to be valuable as an additional tool for reading and presentation of head and neck angiographic studies, especially since the algorithm runs fully automatic. References [1] Hollingworth W, Nathens AB, Kanne JP, et al. The diagnostic accuracy of computed tomography angiography for traumatic or atherosclerotic lesions of the carotid and vertebral arteries: a systematic review. Eur J Radiol 2003;48(1): [2] Josephson SA, Bryant SO, Mak HK, et al. Evaluation of carotid stenosis using CT angiography in the initial evaluation of stroke and TIA. Neurology 2004;63(3): [3] Koelemay MJ, Nederkoorn PJ, Reitsma JB, et al. Systematic review of computed tomographic angiography for assessment of carotid artery disease. Stroke 2004;35(10): [4] Lell M, Fellner C, Baum U, et al. Evaluation of carotid artery stenosis with multisection CT and MR imaging: influence of imaging modality and postprocessing. AJNR Am J Neuroradiol 2007;28(1): [5] Loeckx D, Coudyzer W, Maes F, et al. Nonrigid registration for subtraction CT angiography applied to the carotids and cranial arteries. Acad Radiol 2007;14(12): [6] Tomandl BF, Hammen T, Klotz E, et al. Bone-subtraction CT angiography for the evaluation of intracranial aneurysms. AJNR Am J Neuroradiol 2006;27(1):55 9. [7] Lell M, Anders K, Klotz E, et al. Clinical evaluation of bone-subtraction CT angiography (BSCTA) in head and neck imaging. Eur Radiol 2006;16(4): [8] Flohr TG, McCollough CH, Bruder H, et al. First performance evaluation of a dual-source CT (DSCT) system. Eur Radiol 2006;16(2): [9] Johnson TR, Krauss B, Sedlmair M, et al. Material differentiation by dual energy CT: initial experience. Eur Radiol 2007;17(6): [10] Meyer BC, Werncke T, Hopfenmuller W, et al. Dual energy CT of peripheral arteries: effect of automatic bone and plaque removal on image quality and grading of stenoses. Eur J Radiol 2008;68(3): [11] Gratama van Andel HA, Venema HW, Streekstra GJ, et al. Removal of bone in CT angiography by multiscale matched mask bone elimination. Med Phys 2007;34(10): [12] Lell MM, Ditt H, Panknin C, et al. Cervical CT angiography comparing routine noncontrast and a late venous scan as masks for automated bone subtraction: feasibility study and examination of the influence of patient motion on image quality. Invest Radiol 2008;43(1): [13] Lell MM, Ditt H, Panknin C, et al. Bone-subtraction CT angiography: evaluation of two different fully automated image-registration procedures for interscan motion compensation. AJNR Am J Neuroradiol 2007;28(7): [14] van Straten M, Venema HW, Streekstra GJ, et al. Removal of bone in CT angiography of the cervical arteries by piecewise matched mask bone elimination. Med Phys 2004;31(10): [15] Silvennoinen HM, Ikonen S, Soinne L, et al. CT angiographic analysis of carotid artery stenosis: comparison of manual assessment, semiautomatic vessel analysis, and digital subtraction angiography. AJNR Am J Neuroradiol 2007;28(1):

7 C. Thomas et al. / European Journal of Radiology 76 (2010) [16] Zhang Z, Berg MH, Ikonen AE, et al. Carotid artery stenosis: reproducibility of automated 3D CT angiography analysis method. Eur Radiol 2004;14(4): [17] Busch S, Johnson TR, Nikolaou K, et al. Visual and automatic grading of coronary artery stenoses with 64-slice CT angiography in reference to invasive angiography. Eur Radiol 2007;17(6): [18] Uotani K, Watanabe Y, Higashi M, et al. Dual-energy CT head bone and hard plaque removal for quantification of calcified carotid stenosis: utility and comparison with digital subtraction angiography. Eur Radiol [19] Blum MB, Schmook M, Schernthaner R, et al. Quantification and detectability of in-stent stenosis with CT angiography and MR angiography in arterial stents in vitro. AJR Am J Roentgenol 2007;189(5): [20] Carbone I, Francone M, Algeri E, et al. Non-invasive evaluation of coronary artery stent patency with retrospectively ECG-gated 64-slice CT angiography. Eur Radiol 2008;18(2): [21] Maintz D, Burg MC, Seifarth H, et al. Update on multidetector coronary CT angiography of coronary stents: in vitro evaluation of 29 different stent types with dual-source CT. Eur Radiol [22] Vega Higuera F, Sauber N, Tomandl B, et al. Enhanced 3D-visualization of intracranial aneurysms involving the skull base. Proc Med Image Computing Comput Assist Intervention 2003: [23] Abrahams JM, Saha PK, Hurst RW, et al. Three-dimensional bone-free rendering of the cerebral circulation by use of computed tomographic angiography and fuzzy connectedness. Neurosurgery 2002;51(1):264 8 [discussion 8 9].