Automatic Bone Removal Technique in Whole-Body Dual-Energy CT Angiography: Performance and Image Quality

Transcription

1 Vascular and Interventional Radiology Original Research Schulz et al. Bone Removal in CT Angiography Vascular and Interventional Radiology Original Research Boris Schulz 1 Katharina Kuehling Wolfgang Kromen Petra Siebenhandl Matthias Josef Kerl Thomas Josef Vogl Ralf Bauer Schulz B, Kuehling K, Kromen W, et al. Keywords: bone removal, CT angiography, dual-energy CT, image quality DOI: /AJR Received May 2, 2012; accepted without revision May 8, R. Bauer and M. Kerl are consultants for Siemens Healthcare. 1 All authors: Department of Diagnostic and Interventional Radiology, Clinic of the Goethe University, Haus 23 C UG, Theodor-Stern-Kai 7, Frankfurt, Germany. Address correspondence to B. Schulz (boris.schell@googl .com). WEB This is a Web exclusive article. AJR 2012; 199:W646 W X/12/1995 W646 American Roentgen Ray Society Automatic Bone Removal Technique in Whole-Body Dual-Energy CT Angiography: Performance and Image Quality OBJECTIVE. The purpose of this study was to evaluate the efficiency of automatic bone removal in dual-energy CT angiography (CTA) of the trunk. SUBJECTS AND METHODS. Nineteen patients underwent dual-energy CTA of the trunk (tube A, 140 kv; tube B, 100 kv). In addition to the dual-energy dataset, an image equivalent to that of a standard 120-kV single-energy examination was generated with both tubes. Automated bone segmentation was performed on both datasets, and the results were analyzed. The time required for and subjective image quality of the maximum intensity projections (MIPs) generated were evaluated. RESULTS. Errors in bone segmentation were found for 1.5% of bones on dual-energy images and 12.4% of bones on single-energy images (p < 0.01). The most important differences were found in the rib cage, sternum, and pelvis. The times required for postprocessing of MIPs were similar for the dual-energy (113.5 seconds) and single-energy (106.8 seconds) techniques. The subjective image quality of the arteries was considered better for dual-energy CTA (4.5 points) than for single-energy CTA (4.1 points) owing to false cutoff of vessels during the bone removal process on the single-energy images (p = 0.026). CONCLUSION. For CTA of the trunk, the dual-energy postprocessing capabilities for 3D visualization are superior to the threshold-based bone removal of single-energy CT. Dualenergy CTA can generate boneless MIP images of substantial quality. C T angiography (CTA) is a highly efficient imaging modality for evaluation of the arterial system and is comparable to conventional angiography [1 5]. Unlike conventional projection angiography and digital subtraction angiography, however, CTA is easy to perform and minimally invasive for examining large volumes within seconds with submillimeter spatial resolution. Furthermore, datasets with isotropic voxels acquired with MDCT can be quickly postprocessed into 3D reconstructions, such as volume-rendered and maximum-intensity-projection (MIP) images that can be used for comprehending the topographic context of pathologic findings and to plan interventional and surgical therapies. Software-based tools can be used on postprocessing datasets to automatically separate bones that obstruct or overlie relevant FOVs to generate a boneless 3D rendered image. However, these tools rely on particular attenuation values to differentiate bone from soft tissue, and errors can occur that result in incomplete bone removal and false removal of vessel parts. The combined dataset acquired with two tubes at different voltage levels can be used for tissue discrimination to a certain level, of iodine content in particular [6 8]. The potential for iodine separation in dual-energy CT (DECT) based on spectral information on iodine and bone has been evaluated for automated bone segmentation in the head and neck, and DECT has been found superior to threshold-based single-energy techniques [9 13]. However, the performance of bone subtraction and 3D image quality with DECT in comparison with single-energy CT is yet unclear for whole-body CTA. An important anatomic difference in comparison with the head and neck region that may influence the results has to be considered: The increased body diameter of the abdomen can absorb excessive lower-energy x-rays emitted from the low-kilovoltage tube. Hence, the clear iodine separation necessary for exact tissue discrimination can be impaired. The purpose of this study was to analyze the performance of automated bone subtraction based on spectral information in whole-body dual-energy CTA W646 AJR:199, November 2012

2 Bone Removal in CT Angiography in terms of objective and subjective image quality and timely acquisition. Subjects and Methods Patient Enrollment and Examination Technique Nineteen patients (10 men, nine women; mean age, 64.0 years; range, years) examined between December 2009 and February 2010 were included in this study. The indications for CTA were evaluation of aneurysms, dissection, and postoperative follow-up. Informed consent was obtained from all patients or their legal representatives after the nature of the procedure had been fully explained. Dual-energy CTA was performed with a dual-source 128-MDCT scanner with dual-energy capability (Somatom Definition Flash, Siemens Healthcare). Data were acquired in the craniocaudal direction with the following scan parameters: 140 kv with tin filter; 90-mAs reference tube current time product for tube A; 100 kv and 382 mas for tube B; automated tube current control activated; rotation time, 0.5 second; collimation, 0.6 mm; 128 slices per rotation ( slices in flying focal spot technique); pitch, 1.2. Thinslice transverse images with an FOV of 38 cm were calculated from raw data with a thickness of 1.5 mm and an increment of 1.0 mm. A total of three image sets per examination were reconstructed with the scanner. Two sets of images each contained the separate data from tube A and tube B. The third set combined 60% of the low-kilovoltage information and 40% of the highkilovoltage information for generation of a standard 120-kV single-energy CT image. All images were reconstructed with a medium soft kernel (D26 for the DECT dataset, B26 for the combined tube dataset). All patients received 90 ml of nonionic iodinated contrast material (iomeprol, Imeron 400, Bracco) administered with a power injector (Injektron CT2, Medtron) at a rate of 5 ml/s Error Rate (%) Dual energy Single energy Ribs Clavicles Completely Missed Vertebrae Vertebral Body Missed <50% Vertebral Body Missed >50% and followed by a chaser bolus of 40 ml of saline solution at the same flow rate. Bolus-triggered acquisition timing was used to initiate CTA. For this purpose, the region of interest was placed in the ascending aorta. CTA was started with a delay of 5 seconds after the enhancement curve reached 140 HU. Patients were asked to hold their breath in deep inspiration during scanning of the thoracic region and to continue breathing when the CT device reached the mid abdomen. Postprocessing Both the DECT and the virtual 120-kV single-energy CT datasets were transferred to a multimodality workstation (Syngo MMWP, version VA 21A, Siemens Healthcare) for postprocessing. Bone segmentation based on the spectral dual-energy information was performed with the dual-energy application. Automated bone removal from the combined dataset was performed with the In Space tool. For both methods of bone removal, the standard default settings of the software were used without further manipulation. Each of the postprocessed datasets was converted into 36 MIP images rotated 360 clockwise in 10 steps. Performance Analysis of Bone Subtraction and Postprocessing To evaluate the accuracy of automated bone removal calculations in DECT and single-energy datasets, osseous structures in the scan range were analyzed individually. The clavicles, ribs, and vertebrae were counted separately, and the pelvis and sternum were counted as single bones because of their fused anatomy. Vertebrae were evaluated more precisely for completeness of segmentation (fully segmented, > 50% segmented, < 50% segmented, completely omitted). The total time necessary for postprocessing was noted for each DECT and single-energy CT image set. Time measurements started with the beginning of the bone segmentation process and ended when Sternum Pelvic Bones Fig. 1 Graph shows percentage of bones not deleted in automated bone subtraction during wholebody CT angiography. the 360 MIP rotation was complete. The initial segmentation results were not further manipulated. The intention was to assess the quality achievable with the automated bone removal software in one mouse click. Three radiologists blinded to the bone removal technique individually and independently evaluated each MIP image set at the same multimodality workstation. Subjective impression of the MIP images regarding the texture quality and ability to identify the vessels against the surrounding tissue was assessed on a 5-point Likert scale (1, unacceptable; 2, moderate; 3, average; 4, good; 5, excellent image quality). For detailed evaluation of vessel imaging, the arterial system was subdivided into 18 segments: aorta, vertebral arteries, subclavian arteries, common carotid arteries, mammary arteries, left and right hepatic arteries, gastroduodenal artery, superior mesenteric artery, renal artery, external and internal iliac arteries, and common femoral arteries. Statistical Analysis Descriptive and analytical statistical analyses were performed, and frequencies were calculated. Normal distribution was excluded with the Kolmogorov- Smirnov test. The Mann-Whitney U test was used to compare the differences in means of bone subtraction accuracy, image quality, and duration of both procedures. A CI of 95% (α 0.05) was assumed. Results Efficiency of Automatic Bone Segmentation Process A total of 855 bones each for single-energy and DECT examinations were evaluated with both methods of automatic bone removal. False segmentations were seen in 1.5% of bones with the dual-energy method and 12.4% of bones with the virtual 120-kV thresholdbased bone removal approach (p < 0.01). Rib removal missed 43 ribs in the single-energy approach, and dual-energy removal did not separate 3 of 456 bones (0.7%). The sternum was not deleted in four and the pelvis in five cases with single-energy technique, but dualenergy bone removal did not fail for either of these structures. Sixteen vertebral bodies were not identified correctly with single-energy CT bone removal; a visible vertebra less than 50% of the original size was overlooked with this approach. Dual-energy bone removal did not delete vertebrae completely in eight cases. Figure 1 shows the error rates by bone for single- and dual-energy bone removal. Subjective Image Quality Rating Representative images and the ratings of subjective image quality of the vessels displayed with single-energy CT and DECT are shown AJR:199, November 2012 W647

3 Schulz et al. Fig year-old man with implanted stent graft in the descending thoracic aorta. A C, Frontal maximum-intensityprojection CT angiograms without bone removal (A), with attenuation-based automatic bone removal (B), and spectral automatic bone removal (dualenergy mode) (C) show implanted stent-graft in thoracic aorta. Bones (arrows, B) were missed on attenuation-based image. in Figures 2 and 3. Considering all evaluated vessels, on a 5-point Likert scale, DECT had an average image quality score of 4.5 (SD, 0.5) and single-energy CT an average score of 4.1 (SD, 0.7). This difference was statistically significant (p = 0.026). Although MIP is a standardized display method that should not differ between techniques, automatic bone subtraction deleted parts of vessels in single-energy CT more often than in DECT. Figure 4 illustrates the problem of false-negative removal of arteries with the thresholdbased bone removal function. Time Frame and Radiation Dosage The total time from the start of the segmentation process to finalizing the rotating MIP images was (SD, 11.7) seconds (range, seconds) for the dual-energy method and (SD, 11.7) seconds (range, seconds) for the virtual 120-kV thresholdbased method. The difference was not statistically significant (p = 0.28). According to the automatic dose calculations performed with the CT scanner after the examination, for dual-energy CTA the average volume CT dose index was 9.9 (SD, 1.6) mgy(range, mgy), and the average dose-length product (DLP) was (SD, 133.7) mgy cm (range, mgy cm). With the appropriate conversion factor (k = 0.015), effective dose can be calculated with the DLP (DLP k) [14]. The average calculated effective dose was 10.5 msv (minimum, 7.9 msv; maximum, 12.9 msv). A Fig. 3 Graph shows results of subjective image quality assessment of individual arteries with maximum-intensityprojection technique. SMA = superior mesenteric artery. Image Quality Score Aorta Vertebral Arteries Subclavian Arteries B Discussion Material differentiation, as of iodinated contrast material and bone, based on the spectral behavior of the materials is the key feature of DECT. This principle is used to differentiate contrast agent filled vessels from surrounding bony structures and to generate 3D rendered images that are free of superimposed tissue and bones that would obscure vascular anatomic and pathologic features. Data shown as volume-rendered images can help clinicians achieve an in situ point of view of pathologic conditions and understand the topographic context. Another benefit of these images is to present a fluoroscopy-like set of images to provide a familiar image impression to attending physicians, who are used to conventional angiographic images. The dual-energy technique is based on the characteristic attenuation behavior of iodine and calcified bone at high and low photon energies emitted by the two tubes, which are run with different tube potentials. However, the traditional threshold-based approach whereby vessels are differentiated from surrounding bone mainly on the basis of attenuation on single-energy CT images is currently the most widely used method. In this study we compared the quality of bone removal with an algorithm based on dual-energy information and the quality achieved with the conventional thresholdbased single-energy approach. Dual energy Single energy Carotid Arteries Mammary Arteries Left Hepatic Artery Right Hepatic Artery Gastroduodenal Artery SMA Renal Artery Internal Iliac Arteries External Iliac Arteries Common Femoral Arteries C W648 AJR:199, November 2012

4 Bone Removal in CT Angiography The overall efficiency of automatic bone removal was excellent in dual-energy CTA with correct segmentation of 96.8% in all examined anatomic areas. The greatest difference from results with the standard threshold-based single-energy approach was found in areas close to arteries and bones such as the pelvis. Where vessels take a close course along or through bony structures, error rates with dual-energy bone removal were up to seven times lower than those with the virtual 120-kV single-energy bone removal algorithm. These findings are in accordance with data in the literature that show clear superiority of dual-energy technique in head and neck angiography [10, 13]. The authors of those reports, however, mentioned problems with accurate segmentation of vertebral arteries, especially with but not limited to virtual single-energy CT datasets (the scores for proximal vertebral arteries in our study were 4.3 for the dual-energy technique and 2.6 for the virtual 120-kV technique). The possible explanation for the difference in segmentation quality may be the similar attenuations of bone and contrast material, which can cause difficulties for the software in differentiating vessel and bone, particularly if the course of the arteries is close to osseous structures [12, 15]. Bone segmentation in single-energy CT datasets relies solely on algorithms that differentiate bone cortex and marrow from iodinated contrast material on the basis of x-ray attenuation. Although the single-energy approach showed overall moderately good performance with correct segmentation in 75% of cases, the fact that approximately every fourth bone was missed or incompletely removed shows the inferiority A Fig year-old man for preoperative assessment of pelvic arteries (heart valve replacement). A, CT image obtained with single-energy threshold-based bone removal shows false-negative complete occlusion of left internal iliac artery (arrowhead). On right side of pelvis, short segmented occlusions of smaller arteries are visible false segmentations (arrow). B, Dual-energy CT image obtained with spectral bone removal shows neither of the apparent pathologic findings in A. of the classic approach in comparison with DECT. Brockmann et al. [16] and Meyer et al. [17] compared dual-energy CTA with digital subtraction angiography of the lower extremities and assessed excellent accuracy of the vessels on MIP images after automated bone removal. Regarding the subjective rating of the quality of the display of the vessels, the two techniques had similar ratings with a slight but statistically significant advantage for MIPs rendered with the dual-energy algorithm. In principle, our results show that both single-energy and dual-energy datasets are suitable for visualization of MIP images. However, single-energy CT images suffered more from false cutoff of vessels, leading to an impression of stenosis or occlusion. Therefore, the single-energy technique should be used only with special caution. In addition to ease of use, a requirement for new imaging techniques is seamless integration into clinical workflow. Because expensive, time-consuming, and complicated procedures may be avoided in daily routine, we analyzed the time needed to generate MIP images with dual energy based and virtual single energy based algorithms. The total amount of time from the start of segmentation to MIP reconstruction differed only slightly between techniques (dual energy, 115 seconds; virtual single energy, 107 seconds), and the effort increased with the number of images in a dataset. We believe that a total duration of approximately 2 minutes for generating bone-free 3D images is reasonable in daily routine, whether the processing is performed by a technician or a radiologist. Brockmann et al. [16] reported contrary results regarding time to MIP (dual B energy, 79 seconds; virtual single energy, 155 seconds). The difference may have been due to different computing powers of the postprocessing workstations used. The dose parameters of the dual-energy CTA protocol were a set volume CT dose index equivalent to our single-energy protocol in use at the time of the study (120 kv, 180 mas) because we believe the potential benefit of dual energy based iodine separation does not legitimate increased dose. Our results indicate that dose-equivalent settings are sufficient for accurate tissue discrimination. Johnson et al. [7] also reported that there was no increased exposure with the use of dual-energy protocols per se, yet the performance of dual-energy CTA is still unclear under decreased dosage settings. Regarding radiation hygiene, we believe that it is necessary to analyze the performance of the dual energy based bone removal algorithm with lower dose settings in the future because there is a clear trend to low-kilovoltage CTA with the benefits of lower dose and increased iodine signal. Advanced DECT techniques, such as enhanced vessel attenuation (optimum contrast) may be a valuable tool in these situations [18]. A further feature of dual-energy CTA evaluated positively by Thomas et al. [9] is the additional automated detection and removal of calcified plaque that can obliterate vessels on MIP images. A problem in both single- and dual-energy CTA is beam-hardening artifacts, which occur especially when high-density objects are present in the scanning. Examples are metal objects, such as pacemakers and dental fillings, and highly concentrated contrast material. Vessel narrowing and complete occlusion can be mimicked, and vascular segments can escape interpretation on the final 3D rendered images; thus false-negative and false-positive diagnoses can be made [11, 15]. Although stent evaluation was not a part of this study, four patients had metal stents implanted in the thoracic aorta. Both bone removal techniques resulted in incomplete subtraction, and thus a pseudofragmented shape of the stent was seen in one of the four cases. The bone removal feature itself is not necessary for reading but rather improves visualization. However, we believe that evaluation of thin-slice multiplanar reconstruction source data is mandatory before reading of 3D images such as MIP and volume-rendered images. Even when dual-energy technique is used for CTA, MIPs should not be used to assess or exclude pathologic conditions in vessels. AJR:199, November 2012 W649

5 Schulz et al. Although digital subtraction angiography still has the advantage of intravascular treatment of stenoses, aneurysms, and other lesions, CTA is a minimally invasive, economically attractive, and widely available technique for analyzing a large anatomic volume at high spatial resolution to guide intervention and surgery. With modern postprocessing techniques, pathologic findings can be visualized within seconds on 3D rendered images. Furthermore, dual-energy CTA is superior to single-energy techniques in terms of automated bone segmentation for whole-body CTA. At equivalent radiation exposure and time investment for postprocessing, dual energy is the ideal technique for generating 3D rendered images. References 1. Zhang LJ, Wu SY, Poon CS, et al. Automatic bone removal dual-energy CT angiography for the evaluation of intracranial aneurysms. J Comput Assist Tomogr 2010; 34: Heuschmid M, Krieger A, Beierlein W, et al. Assessment of peripheral arterial occlusive disease: comparison of multislice-ct angiography (MS- CTA) and intraarterial digital subtraction angiography (IA-DSA). Eur J Med Res 2003; 8: Martin ML, Tay KH, Flak B, et al. Multidetector CT angiography of the aortoiliac system and lower extremities: a prospective comparison with digital subtraction angiography. AJR 2003; 180: Kumamaru KK, Hoppel BE, Mather RT, et al. CT angiography: current technology and clinical use. Radiol Clin North Am 2010; 48: Uotani K, Watanabe Y, Higashi M, et al. Dualenergy CT head bone and hard plaque removal for quantification of calcified carotid stenosis: utility and comparison with digital subtraction angiography. Eur Radiol 2009; 19: Hawkes DJ, Jackson DF, Parker RP. Tissue analysis by dual-energy computed tomography. Br J Radiol 1986; 59: Johnson TR, Krauss B, Sedlmair M, et al. Material differentiation by dual energy CT: initial experience. Eur Radiol 2007; 17: Thieme SF, Johnson TR, Lee C, et al. Dual-energy CT for the assessment of contrast material distribution in the pulmonary parenchyma. AJR 2009; 193: Thomas C, Korn A, Krauss B, et al. Automatic bone and plaque removal using dual energy CT for head and neck angiography: feasibility and initial performance evaluation. Eur J Radiol 2010; 76: Deng K, Liu C, Ma R, et al. Clinical evaluation of dual-energy bone removal in CT angiography of the head and neck: comparison with conventional bone-subtraction CT angiography. Clin Radiol 2009; 64: Lell MM, Kramer M, Klotz E, et al. Carotid computed tomography angiography with automated bone suppression: a comparative study between dual energy and bone subtraction techniques. Invest Radiol 2009; 44: Lell M, Anders K, Klotz E, et al. Clinical evaluation of bone-subtraction CT angiography (BSC- TA) in head and neck imaging. Eur Radiol 2006; 16: Morhard D, Fink C, Graser A, et al. Cervical and cranial computed tomographic angiography with automated bone removal: dual energy computed tomography versus standard computed tomography. Invest Radiol 2009; 44: Bongartz G, Golding SJ, Jurik AG, et al. European guidelines for multislice computed tomography. Luxembourg, Luxembourg: European Commission, Tomandl BF, Hammen T, Klotz E, et al. Bonesubtraction CT angiography for the evaluation of intracranial aneurysms. AJNR 2006; 27: Brockmann C, Jochum S, Sadick M, et al. Dualenergy CT angiography in peripheral arterial occlusive disease. Cardiovasc Intervent Radiol 2009; 32: 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: Holmes DR 3rd, Fletcher JG, Apel A, et al. Evaluation of non-linear blending in dual-energy computed tomography. Eur J Radiol 2008; 68: W650 AJR:199, November 2012