Evaluation of surface and volume rendering in 3D-CT of facial fractures

Transcription

1 (2006) 35, q 2006 The British Institute of Radiology RESEARCH Evaluation of surface and volume rendering in 3D-CT of facial fractures T Rodt*,1, SO Bartling 2, JE Zajaczek 2, MA Vafa 3, T Kapapa 1, O Majdani 4, JK Krauss 1, M Zumkeller 1, H Matthies 5, H Becker 2 and J Kaminsky 6 Departments of 1 Neurosurgery, 2 Neuroradiology, 3 Radiology, 4 Otorhinolaryngology and 5 Medical Informatics, Hannover Medical School, Germany; 6 Department of Neurosurgery, Tuebingen University Medical School, Germany Objectives: Three-dimensional computed tomography (3D-CT) of facial fractures has been reported as beneficial using surface (SR) and volume rendering (VR). There are controversial statements concerning the preferable algorithm. The purpose of this study was to evaluate and compare SR and VR for clinical 3D-CT in facial fractures on an experimental basis. Methods: Multislice CT was obtained in 22 patients with facial fractures using two data acquisition protocols. Five SR and VR post-processing protocols were applied. Five assessors independently evaluated the quality of visualization of the fracture gap and dislocated fragments as well as the overall image quality using a five-point rating scale. The potential benefit of the 3Dimages for radiological diagnosis and presentation was evaluated. The influence of the data acquisition protocol was analysed. Results: SR in general achieved better evaluation scores than VR at corresponding thresholds. Variation of evaluation scores for all criteria was found for SR and VR depending on the segmentation threshold. Apart from the overall image quality no significant influence of the data acquisition technique was found for the evaluated criteria. Conclusions: SR provided sufficient and time efficient means for 3D-visualization of facial fractures in this study. No diagnostic benefit of VR over SR was found. (2006) 35, doi: /dmfr/ Keywords: computed tomography (CT), three-dimensional; facial fractures; images, processing; bones, CT Introduction Current multislice CT technologies allow high resolution imaging in patients with craniofacial trauma. 1 3 Assessment of facial fractures by cross-sectional CT images alone is often difficult due to the complex anatomy and the various possible fracture patterns. 2,4 Furthermore, several classifications describing the fracture patterns exist Dvisualization has been reported to facilitate evaluation of CT data in patients with facial trauma by displaying the spatial relationship of the different anatomical and pathological structures such as fracture gaps and fracture fragments. 4,8,9 Therefore, it can be beneficial for radiological diagnosis, presentation and surgical planning. 10 The most widespread automated post-processing *Correspondence to: Thomas Rodt, MD, Department of Neurosurgery, Hannover University Medical School, Carl-Neuberg-Str. 1, Hannover, Germany; rodt@gmx.de Received 4 July 2005; revised 20 October 2005; accepted 30 October 2005 algorithms for 3D-visualization of CT-data are surface rendering (SR) and volume rendering (VR). 11,12 Controversial statements have been made regarding which of the two algorithms, SR or VR, is superior for 3Dvisualization of fractures in general and for the specific case of facial fractures. 11,13 15 Both algorithms have advantages and disadvantages when used for clinical imaging where image quality as well as post-processing time is of importance. SR reduces the amount of data that has to be rendered and therefore limits the calculation time effort for post-processing. A determined surface within the data volume is displayed by calculation of overlapping polygons. 12 The automated determination or segmentation of the surface is often achieved by thresholding. Other manual and automated segmentation techniques have been described but have disadvantages due to interobserver or intraobserver variability and time effort for the purpose of clinical imaging. 13,16,17 When VR is used the whole amount of data

2 228 Surface and volume rendering in 3D-CT T Rodt et al is taken into account to render a 3D-image by adding the voxel values along a virtual light ray through the volume. Several transfer functions have been described, differing in the way the individual voxels contribute to the final images. By taking the complete amount of data into account VR can show more information but is computationally more demanding than SR. 12 To our knowledge no studies comparing different SRand VR-algorithms for clinical 3D-visualization of facial fractures have been performed thus far. The aim of this retrospective study was to evaluate and compare SR and VR for clinical 3D-visualization of facial fractures on an experimental basis. Material and methods Patients CT datasets of 22 patients with facial fractures were examined retrospectively. 18 patients were male and four were female. The age range was 1 64 years (mean ¼ 30.5 years, SD ¼ 14.4 years). Fractures included the mandible, the maxilla, the frontal bone, the zygomatic arch and the nasal bone. Four patients had multiple facial fractures. CT data was obtained within a maximum of 3 days following trauma. Data acquisition Data acquisition was performed using helical multislice CT. Two different scanners and data acquisition protocols were applied. 15 patients were scanned using the routine low-dose protocol 1 (120 kv, 100 ma, pitch 3, 1.25 mm collimation, no contrast agent) with a LightSpeed QX/i CT (General Electric (GE) Medical Systems, Milwaukee, WI). Seven polytrauma patients were scanned using dataacquisition protocol 2 designed for assessment as part of a whole body scan (120 kv, 175 ma, pitch 3, 1.25 mm collimation, 50 ml of contrast agent were administered in the emergency room) with a Somatom Volume Zoom CT (Siemens, Munich, Germany). In both settings final image reconstruction was achieved by applying a high-resolution algorithm at 1 mm interval, 20 cm field of view (FOV) and a matrix. Post-processing The datasets were transferred to an Ultra10-workstation (Sun Microsystems, Santa Clara, CA) via an internal Dicom 3.0 network. The software Advantage Windows 3.1 (GE) was used for post-processing. 3D-images were created using five post-processing protocols (Table 1). Corresponding views of the 3D-models were created and printed on films (Figures 1 and 2). Evaluation Evaluations were made by five independent assessors who had experience in neuroradiological assessment of facial trauma. Diagnosis was made based on the 2D-images first. No information concerning the patients or the postprocessing protocols was given to the assessors. A five point rating scale was used to evaluate the 3D-images of all 22 patients for the 5 different algorithms (1 ¼ very good, 2 ¼ good, 3 ¼ fair, 4 ¼ sufficient, 5 ¼ insufficient). For each protocol five criteria were assessed (overall image quality, visualization of fracture gaps, visualization of fracture fragments, potential diagnostic benefit and potential benefit for presentation). The term potential benefit for presentation refers to communication of a pathology of an anatomic structure along with the adjacent normal anatomytocolleagues. Statistical analysis Statistical analysis was performed using SPSS 10.0 (SPSS Inc., Chicago, IL). For each criterion assessed the mean of the five evaluations was calculated for all 22 patients. Overall mean and standard deviation were calculated for all patients, and for the two data-acquisition subgroups. A variance analysis with repeated measurements was performed to detect whether a statistically significant difference between the five post-processing protocols existed and to reveal differences between the two dataacquisition subgroups. Hypothesis 1 was that differences between the achieved evaluation values of the different post-processing protocols existed. Hypothesis 2 was that a significant interaction between the evaluation values and the data-acquisition subgroups existed. Results Data-transfer and 3D-visualization of all relevant structures were feasible in all cases. Post-processing took approximately 1 min for SR and 5 min for VR. 3Dvisualization facilitated assessment of the spatial relationship of the fracture and the anatomy. The overall image quality was acceptable in all cases. The SR post-processing protocols 1 and 2 were rated better than the VR post-processing protocols 3, 4 and 5 for all evaluated criteria. The SR protocol 2 with a threshold of 260 HU achieved better overall mean scores than the SR protocol 1 with a threshold of 160 HU for all criteria. The VR protocol 4 with a lower threshold of 160 HU achieved nearly identical overall mean scores to VR protocol 5 with a threshold of 260 HU for all criteria. The VR protocol 3 resulted in the worst overall mean evaluations. The overall mean and standard deviation of the five post-processing protocols are displayed in Figure 3. Statistical analysis showed that the differences of the five algorithms were highly significant for all criteria (P, ). Hypothesis 1 was corroborated. Statistical analysis of the relationship between the dataacquisition technique and the evaluation scores showed no significant interaction for criteria 2 5 (P ). A significant interaction (P ¼ 0.004) between the dataacquisition technique and the evaluation scores was found for criterion 1 (overall image quality). The overall mean and standard deviation are compared with the mean and standard deviation of the two different data-acquisition techniques for criterion 1 in Figure 4. The evaluation scores for overall image quality were more unfavourable when data-acquisition was performed using the trauma

3 Surface and volume rendering in 3D-CT TRodtet al 229 Table 1 Parameters of the evaluated post-processing protocols Protocol SR/VR Parameters 1 SR lower threshold 160 HU, Select Object 2 SR lower threshold 260 HU, Select Object 3 VR linear upramp curve from 0% opacity and 95 HU to 70% opacity and 1500 HU, 70% opacity from 1500 HU on 4 VR 100% opacity from 160 HU on 5 VR 100% opacity from 260 HU on SR, surface rendering; VR, volume rendering protocol. Furthermore, the advantage of SR post-processing over VR post-processing was smaller as compared with the routine data-acquisition. Hypothesis 2 was not corroborated for the criteria 2 5, but was corroborated for the criterion 1. Discussion The benefit of 3D-visualization for radiological assessment, surgical planning and presentation has been reported in several studies. 4,8,10 Since a retrospective analysis was performed in the present study only the potential diagnostic benefit could be evaluated. It has to be mentioned that the whole information is present in the complete set of 2D-images, thus correct diagnosis should be possible in all cases based on 2D-images alone. However, 3D-visualization as an additional diagnostic tool facilitates radiological diagnosis by efficiently presenting information on the spatial relationships in a structured and condensed way. 4,8 In the present study only patients with facial fractures were examined. With regard to the distribution of sex and age of the patients, our group represents a typical sample of facial fractures. 18 There are controversial statements on the preferable algorithm for 3D-visualization of fractures using either SR or VR. 11,13 A study published by Udupa et al comparing the algorithms on a technical basis found that SR had some advantages as compared with VR concerning visualization of thin bones, sutures and fractures. 13 Furthermore, computational and storage resources were included as criteria in this study. Other studies that compared the algorithms as part of clinical examinations concluded that VR was the preferable algorithm. 14,15 These studies, however, did not focus on the specific subgroup of patients with facial fractures. The present study aimed at a comparison of SR and VR for clinical application of 3D- CT in patients with facial fractures. For the protocols 1 and 2 (SR), as well as for 4 and 5 (VR) thresholds of 160 HU and 260 HU were chosen as these or similar values were reported to be suitable for 3D-visualization of bony structures. Protocol 3 was included as it was offered as a commercial preset in the software. On a theoretical basis a VR transfer function using an ascending opacity is not suitable for 3D-visualization of facial fractures as Figure 1 Patient with depressed fracture of the left maxilla. (A) Axial multislice CT, (B) 3D-visualization using surface rendering (SR) post-processing protocol 1, (C) 3D-visualization using SR post-processing protocol 2, (D) 3D-visualization using volume rendering (VR) post-processing protocol 3, (E) 3D-visualization using VR post-processing protocol 4, (F) 3D-visualization using VR post-processing protocol 5. SR achieved better evaluation values compared with VR for criteria 2 5. Data-acquisition was performed using protocol 2, overall image quality in some areas was better using VR

4 230 Surface and volume rendering in 3D-CT T Rodt et al Figure 2 Patient with tripod-fracture on the right side. (A) Axial multislice CT, (B) 3D-visualization using surface rendering (SR) post-processing protocol 1, (C) 3D-visualization using SR post-processing protocol 2, (D) 3D-visualization using volume rendering (VR) post-processing protocol 3, (E) 3D-visualization using VR post-processing protocol 4, (F) 3D-visualization using VR post-processing protocol 5. SR achieved better evaluation values for all criteria surface-information can be reduced due to an opacity smaller than 100%. Due to the short post-processing time all protocols compared in this study allow a routine clinical application of 3D-visualization in patients who underwent facial trauma. The two data-acquisition protocols used both resulted in good 2D-image quality. As indicated above, data-acquisition protocol 2 was performed as part of a whole body scan. Thus, the patients were exposed to a higher effective radiation dose. Due to hardware attached to the polytrauma patients more beam hardening artefacts occurred. For clinical imaging, 3D-visualization techniques should be chosen taking into account both diagnostic benefit and time/cost efficiency. With regard to these requests the results of this study imply that the tendency to provide only more complex algorithms as part of commercial software packages has to be viewed critically. Apart from the significant interaction between the data-acquisition technique and the ratings for the overall image quality as described above, no other differences were found. The less favourable ratings for overall image quality with the trauma protocol could be due to the additional hardware Figure 3 Mean evaluation values and standard deviation of all 22 patients for the 5 evaluated post-processing protocols and 5 evaluated criteria. Evaluation was performed by 5 assessors that had experience in assessing facial trauma using a rating scale (1 ¼ very good, 2 ¼ good, 3 ¼ fair, 4 ¼ sufficient, 5 ¼ insufficient)

5 Surface and volume rendering in 3D-CT TRodtet al 231 Figure 4 Mean evaluation values and standard deviation for the overall image quality of all 22 patients, and separately of the patients that underwent data-acquisition protocol 1 and data acquisition protocol 2. Statistical analysis showed a significant interaction of the data-acquisition technique and the achieved mean evaluation value. Volume rendering (VR) evaluation values were still worse than surface rendering (SR) evaluation values but to a smaller extent compared with patients where CT was obtained using protocol 1 attached to the patients causing artefacts and also to the contrast media which were given. 1 In these cases VR ratings were worse than SR ratings, but to a much smaller extent in comparison with the routine clinical data acquisition. In conclusion, in this study the SR algorithm provided sufficient and time efficient means for 3D-visualization of facial fractures. No diagnostic benefit of the VR algorithm was found compensating for the additional time effort in clinical 3D-visualization. References 1. Dammert S, Funke M, Merten HA, Obernauer S, Grabbe E. Multislice helical CT (MSCT) for mid-facial trauma: optimization of parameters for scanning and reconstruction. Rofo 2002; 174: Philipp MO, Funovics MA, Mann FA, Herneth AM, Fuchsjaeger MH, Grabenwoeger F, et al. Four-channel multidetector CT in facial fractures: do we need mm collimation? AJR Am J Roentgenol 2003; 180: Rosenthal E, Quint DJ, Johns M, Peterson B, Hoeffner E. Diagnostic maxillofacial coronal images reformatted from helically acquired thinsection axial CT data. AJR Am J Roentgenol 2000; 175: Dempf R, Hausamen JE. Fractures of the facial skull. Unfallchirurg 2000; 103: Donat TL, Endress C, Mathog RH. Facial fracture classification according to skeletal support mechanisms. Arch Otolaryngol Head Neck Surg 1998; 124: Kos M, Luczak K, Godzinski J, Rapala M, Klempous J. Midfacial fractures in children. Eur J Pediatr Surg 2002; 12: Buitrago-Tellez CH, Schilli W, Bohnert M, Alt K, Kimmig M. A comprehensive classification of craniofacial fractures: postmortem and clinical studies with two- and three-dimensional computed tomography. Injury 2002; 33: Joos U, Piffko J, Meyer U. Treatment of frontobasal trauma and polytrauma. Mund Kiefer Gesichtschir 2001; 5: Wang G, Vannier MW. Stair-step artifacts in three-dimensional helical CT: an experimental study. Radiology 1994; 191: Mankovich NJ, Samson D, Pratt W, Lew D. Beumer 3rd J. Surgical planning using three-dimensional imaging and computer modeling. Otolaryngol Clin North Am 1994; 27: Rieker O, Mildenberger P, Rudig L, Schweden F, Thelen M. 3D CT of fractures: comparison of volume and surface reconstruction. Rofo 1998; 169: Calhoun PS, Kuszyk BS, Heath DG, Carley JC, Fishman EK. Threedimensional volume rendering of spiral CT data: theory and method. Radiographics 1999; 19: Udupa JK, Hung HM, Chuang KS. Surface and volume rendering in three-dimensional imaging: a comparison. J Digit Imaging 1991; 4: Kuszyk BS, Heath DG, Bliss DF, Fishman EK. Skeletal 3-D CT: advantages of volume rendering over surface rendering. Skeletal Radiol 1996; 25: Benateau H, Chevallier E, Hamon M, Edy E, Keswani R, Labbe D, et al. The three-dimensional spiral scanner and volume rendering technique: importance in craniofacial traumatology and reconstructive surgery. Rev Stomatol Chir Maxillofac 2002; 103: Kikinis R, Shenton ME, Gerig G, Martin J, Anderson M, Metcalf D, et al. Routine quantitative analysis of brain and cerebrospinal fluid spaces with MR imaging. J Magn Reson Imaging 1992; 2: Shin H, Stamm G, Hogemann D, Galanski M. Basic principles of data acquisition and data processing for construction of high quality virtual models. Radiologe 2000; 40: Meyer U, Benthaus S, Du Chesne A, Wannhof H, Zollner B, Joos U. Examining patients with facial skull fractures from an etiological and legal perspective. Mund Kiefer Gesichtschir 1999; 3: