UvA-DARE ( Digital Academic Repository ) Image processing in vascular computed tomography

For clear visualization of vessels in CT angiography (CTA) images of the head and neck using maximum intensity projection (MIP) or volume rendering (VR) bone has to be removed. In the past we presented a fully automatic method to mask the bone (matched mask bone elimination, MMBE) for this purpose. A drawback is that vessels adjacent to bone may be partly masked as well. We propose a modification, multiscale MMBE, which reduces this problem by using images at two scales: a higher resolution than usual is used for image processing, and a lower resolution to which the processed images are transformed for use in the diagnostic process. A higher in-plane resolution is obtained by the use of a sharper reconstruction kernel. The out-of-plane resolution is improved by deconvolution or by scanning with narrower collimation. The quality of the mask that is used to remove bone is improved by using images at both scales. After masking, the desired resolution for the normal clinical use of the images is obtained by blurring with Gaussian kernels of appropriate widths. Both methods (multiscale and original) were compared in a phantom study and with clinical CTA data sets. With the multiscale approach the width of the strip of soft tissue adjacent to the bone that is masked can be reduced from 1.0 mm to 0.2 mm without reducing the quality of the bone removal. The clinical examples show that vessels adjacent to bone are less affected and therefore better visible. Images processed with multiscale MMBE have a slightly higher noise level or slightly reduced resolution compared with images processed by the original method and the reconstruction and processing time is also somewhat increased. Nevertheless multiscale MMBE offers a way to remove bone automatically from CT angiography images without affecting the integrity of the blood vessels. The overall image quality of MIP or VR images is substantially improved relative to images processed with the original MMBE method. 25 Removal of Bone in CTa By mulTiSCale maTChed maSk Bone eliminaTion

Matched mask bone elimination (MMBE) is a relatively new technique to remove bone from CTA source images (CTA-MMBE) in an automatic and user-independent way with little additional radiation dose. [20][21][22] In CTA-MMBE, a second nonenhanced low-dose scan (about a quarter of the radiation dose of a regular CTA) is used to identify bony structures that can subsequently be masked in the CTA scan.
Digital subtraction angiography (DSA) is the gold standard for detection of intracranial aneurysms. Extension of DSA with 3D rotational angiography (3DRA) can further improve detection of intracranial aneurysms that may be obscured by overprojecting vessels. [23][24][25] The advantages of DSA over CTA are superior spatial and contrast resolution, no interference of bony structures, and the possibility to perform direct endovascular interventions. 26,27 However, DSA is an invasive technique with a small but significant risk of neurologic complications, estimated to occur in 0.3%-1.8% of patients. 28,29 The purpose of this study was to determine the diagnostic accuracy of CTA-MMBE for detection of intracranial aneurysms in a large patient population with clinically suspected subarachnoid hemorrhage (SAH) with DSA and 3DRA as reference standards.

Patients
Between January 2004 and February 2006, 108 patients who presented with clinically suspected SAH underwent both CTA-MMBE and DSA for diagnosis of an intracranial aneurysm. Of 108 patients, 102 had SAH confirmed by CT or lumbar puncture. There were 81 women and 27 men with a mean age of 56 years (median, 53 years; range, 19 -92 years). In general, when 1 or more aneurysms were found on CTA or DSA, additional 3DRA was performed for pretreatment planning. Approval from the institutional review board for review of the patient's medical records and images was obtained. Because CTA-MMBE and DSA/3DRA were part of routine clinical practice, no approval was required to perform these imaging techniques in the patient group.

Imaging
Description of the technique of CTA-MMBE has been published previously. 20 Parameter settings were used as found optimal by van Straten et al. 22 Briefly, in MMBE, an additional nonenhanced low-dose spiral CT scan (65 mAs) was used to identify bony structures that were subsequently masked on CTA images (Fig 1A-C). These scans were made on a 4-section spiral CT scanner (MX8000; Philips Medical Systems, Best, the Netherlands or Sensation 4; Siemens Medical Solutions, Erlangen, Germany). We used the following parameters: 120 kV, 250 mAs, 4 ϫ 1 mm detector collimation; pitch of 0.875, section thickness of 1.3 mm, increment of 0.5 mm, 150-mm FOV, 512 ϫ 512 matrix, and reconstruction kernels B (Philips Medical Systems) and H30f (Siemens Medical Solutions). Eighty to 100 mL of nonionic contrast material was injected in a cubital vein at a rate of 4 mL/s. Scanning delay was automatically adjusted by a bolus-tracking technique.
CT and CTA images were sent to a workstation, and bone was removed automatically within 3 minutes by using the MMBE technique. CTA scans with masked bone were further processed in standardized maximum-intensity-projection (MIP) images: 40 images of different viewing angles rotated in a vertical and horizontal axis ( Fig  1D). 8,30 After creation of a volume of interest, additional MIP images were made without venous structures.
DSA and 3DRA were performed by an experienced neuroradiologist on a single-plane angiographic unit (Integris Allura Neuro; Philips Medical Systems). Most angiograms were obtained with the patient under general anesthesia before coiling. Through a 6F catheter positioned in an internal carotid artery (ICA) or vertebral artery, 6-to 8-mL nonionic contrast was injected, and filming was performed in 2-3 projections at a frame rate of 2 per second. For 3DRA, 100 images were acquired during a 240°rotational run in 8 seconds with 15-to 21-mL contrast medium at 3 mL/s. On a dedicated workstation, 3D images were constructed and evaluated. Screen shots in multiple projections of volume-rendered 3D images were stored.

CTA-MMBE.
For the purpose of this study, CTA source images and MIP images were anonymized and evaluated independently by 2 neuroradiologists blinded to clinical data and diagnostic CT and DSA results. CTA-MMBE image quality was rated as "good," "fair," "moderate," or "poor" and the extent of bone removal, as "complete," "near-complete," or "incomplete." "Near-complete" bone removal was defined as the presence of tiny bone remnants or small calcifications (Fig 2A, -B). Bone removal was "incomplete" when large bone remnants were present. Both readers recorded reasons for moderate or poor image quality. Twenty-seven predefined locations, subdivided into 4 subgroups, needed to be observed by each reader (Table  1). These 27 predefined locations were evaluated in all patients, resulting in the observation of 2916 locations in total. For interobserver discrepancies in detection of aneurysms, consensus was reached.
DSA and 3DRA. DSA images in combination with screen shots of 3DRA, if available, were used as reference standards. 26,27 Images were re-evaluated by an experienced interventional neuroradiologist. The observer needed to evaluate all visible locations from the 27 predefined locations, which depended on the number of vessels catheterized. Aneurysm size as measured on 3DRA was recorded. If 3DRA was not available, aneurysm size was estimated from comparison with the ICA (4 mm) or basilar artery (3 mm). Finally, rupture status of the aneurysm, derived from all image techniques, was assessed.

Statistical Analysis
The diagnostic accuracy of CTA-MMBE was calculated for 2188 locations observed with both DSA and CTA. Sensitivity, specificity, positive predictive value (PPV), negative predictive value (NPV), and exact 2-sided 95% confidence intervals (CI) were calculated for aneu- rysm detection. 31 Diagnostic accuracy was also calculated per patient, per aneurysm size, per observed location, and per aneurysm location in 4 subgroups. Interobserver variability with percentages of agreement was calculated with statistics, with Ն 0.80 defined as excellent agreement; ϭ 0.60 -0.79, as good agreement; ϭ 0.40 -0.59, as moderate agreement; ϭ 0.20 -0.39, as fair agreement; and ϭ 0 -0.19, as slight agreement. The McNemar test was used to verify whether significant differences between observers and consensus of CTA-MMBE were present on aneurysm detection.

CTA Evaluations
Overall CTA-MMBE image quality and extent of bone removal evaluation for both observers are listed in Table 2. Reasons for poor or moderate image quality or incomplete bone removal (15 patients) were patient movement during scanning (8 patients) and insufficient arterial contrast combined with overprojecting venous structures (7 patients).
Overall diagnostic performance of CTA-MMBE for detection of intracranial aneurysms is listed in Table 3; diagnostic performance per aneurysm subgroup location, in Table 4; and diagnostic performance according to aneurysm size, in Table  5.
With CTA-MMBE, observer A detected 102 of 117 aneurysms (87%) (79 ruptured, 23 unruptured) and observer B, 103 (88%) (80 ruptured, 23 unruptured). After consensus, 11 aneurysms (9%) (1 ruptured, 10 unruptured) remained undetected. Observer A found 1 false-positive finding and observer B found 5 false-positive findings; after consensus, 3 false-positive findings remained. The false-positive finding of observer A proved to be cavernous sinus enhancement; the 5 false-positive findings of observer B were 3 infundibula (Fig 2C, -D), 1 ICA loop, and 1 local thickening of the pericallosal artery.   Sensitivity for detecting aneurysms Ն3 mm was 0.99, and for aneurysms Ͻ3 mm sensitivity was 0.38. Characteristics of 11 aneurysms undetected after consensus with reasons for missing them are listed in Table 6. Ten of 11 missed aneurysms were Յ2.5 mm and were additional unruptured aneurysms in patients harboring multiple aneurysms (see example in Fig 1D-H). One missed aneurysm was a 4-mm ruptured dissecting posterior inferior cerebellar artery (PICA) aneurysm on a poor-quality CTA-MMBE due to patient movement. Another missed aneurysm was a 2.5-mm ophthalmic aneurysm in direct contact with the anterior clinoid process, which appeared smaller on MIP compared with the nonmasked CTA source images due to the applied MMBE technique and was, therefore, evaluated as an irregularity of the ICA. Most missed aneurysms were located on the ICA (5 of 11), and consequently, sensitivity for detection of aneurysms located on the ICA was significantly lower than that for other subgroup locations (Table 4).
Interobserver agreement per aneurysm location and per patient was excellent ( ϭ 0.92 with 99% agreement per location and ϭ 0.80 with 94% agreement per patient, respectively). Due to the high number of negative aneurysm locations, observer agreement per patient was the most representative value. The McNemar test showed no significant differences between observers for aneurysm detection per patient and per location and for observers and consensus.

Discussion
In this study, we found high sensitivities and specificities of CTA-MMBE for detection of intracranial aneurysms with excellent interobserver agreement. Only 1 small dissecting aneurysm out of 82 ruptured aneurysms was not detected due to poor quality CTA-MMBE by patient movement. Other undetected aneurysms were very small unruptured aneurysms additional to a detected ruptured aneurysm. As in other studies, several shortcomings of CTA-MMBE and MIP images leading to misinterpretation were apparent, such as difficulties in differentiating arterial loops or infundibula from aneurysms ( Fig  2C, -D), cavernous sinus contrast enhancement sometimes simulating or obscuring an aneurysm, and lack of depiction of very small aneurysms, especially located on the ICA. 9,10 Our study design differs in some aspects from earlier published reports comparing CTA and DSA. 2,3,32-34 Because DSA and 3DRA do not necessarily include all vascular territories, analysis of results was performed per predefined anatomic aneurysm location and not per detected aneurysm. To minimize verification bias, we also included patients with no aneurysms found on DSA and CTA. To resemble clinical practice, we did not discard low-quality CTA studies. We used high-quality DSA and 3DRA as a reference standard (most angiograms were obtained in patients under general anesthesia), and con-   sequently, aneurysms as small as 1-2 mm were easily detected ( Fig 1G). This standard may partly explain the relatively low sensitivity in this study for these very small intracranial aneurysms. Because our study design differed in analysis and calculation methods, comparison with other studies is of limited value. For instance, in another study with a comparable number of patients, higher sensitivity for the detection of aneurysms was reported (0.95 versus 0.91), but small aneurysms located on the ICA were excluded. 4 In general, results of this study are in concordance with 2 meta-analyses. 26,27 One could suspect that the relatively low sensitivity for small aneurysms could be attributed to the MMBE technique. However, of 11 missed aneurysms, only one 2.5-mm (ophthalmic) aneurysm was directly adjacent to a bony structure. This aneurysm appeared smaller on MIP compared with the nonmasked CTA source images due to the applied MMBE technique. All other missed aneurysms were not directly adjacent to bone, so MMBE could be excluded as a cause of missing them.
Our results and those of others suggest that complete DSA is no longer mandatory as a diagnostic tool in patients with good-quality CTA. Only in patients with confirmed SAH and poor-quality CTAs should complete diagnostic DSA be performed. Detection of a ruptured aneurysm with CTA can be followed by selective DSA of the vessel harboring the aneurysm before endovascular treatment, thereby reducing complication risk and procedural time. If this strategy were applied to the present study population, only very small additional unruptured aneurysms would go undetected. The rupture risk of very small additional aneurysms is extremely low. 35 Because undetected small aneurysms may grow with time, however, a follow-up strategy in patients with coiled intracranial aneurysms should be DSA of the vessel harboring the aneurysm after 6 months and MRA at a later date, for example after 12-18 months and possibly yearly thereafter. With this strategy, growing initially undetected small aneurysms or de novo aneurysms will be detected in a timely manner. Patients with clipped aneurysms can be followed with CTA, especially when titanium clips were used. [36][37][38] Manual bone editing in CTA is time-consuming (approximately 20 minutes) and user-dependent and requires knowl-edge of vascular anatomy. 8,18 In contrast, MMBE for bone removal is fully automatic and user-independent. MMBE removed bone adequately in all patients with good or fair quality CTA examinations with only insignificant small remnants of bone or calcifications (Fig 2 A, -B). When CTA examinations were of moderate or poor quality, CTA-MMBE and MIP images were also of lower quality. The main causes for reduced CTA image quality were patient movement during scanning, insufficient arterial contrast, and overprojection of veins. In case of patient movement during scanning, the incomplete bone removal was restricted to section positions at which movement occurred.
The question of whether MMBE has additional value in aneurysm detection remains unanswered in this study because we did not compare CTA-MMBE with a standard CTA technique. Subjectively however, MMBE offers easier and quicker viewing of reconstructed images without hindering bony structures. Average reading time per case was approximately 10 minutes, which is shorter than other CTA reading times reported (15-20 minutes). 39,40 The use of 16-or 64-section CT scanners with shorter scanning times, thinner sections, and volume-rendered image display will improve the image quality of CTA. This may lead to higher sensitivity for detection of very small aneurysms and improved evaluation of aneurysm characteristics to assess the mode of treatment (surgical or endovascular).

Conclusion
CTA-MMBE is an accurate imaging technique for detection of intracranial aneurysms, allowing rapid aneurysm visualization on MIP images in any projection without overprojecting bone in a fully automatic and operator-independent way. CTA-MMBE has limited sensitivity in detecting aneurysms Ͻ3 mm with the use of 4-section CT scanners. Our data suggest that DSA and 3DRA can be limited to the vessel harboring the ruptured aneurysm before endovascular treatment, after detection of a ruptured aneurysm with CTA.