BJR
HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Kickuth, R
Right arrow Articles by Kirchner, J
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Kickuth, R
Right arrow Articles by Kirchner, J
British Journal of Radiology 75 (2002),422-427 © 2002 The British Institute of Radiology

Full Paper

Stereoscopic 3D CT vs standard 3D CT in the classification of acetabular fractures: an experimental study

R Kickuth, MD1, G Hartung, MD2, U Laufer, MD1, C Gruening, MD2, C Stueckle, MD1, D Liermann, MD1 and J Kirchner, MD1

Departments of 1 Radiology and 2 Surgery, Marienhospital Herne, University of Bochum, Hoelkeskampring 40, D-44625 Herne, Germany


   Abstract
Top
 Abstract
Introduction
 Materials and methods
 Results
 Discussion
 References
 
The accuracy of stereoscopic and standard three-dimensional (3D) CT in the classification of acetabular fractures was compared. A receiver operating characteristic (ROC) analysis was performed by two radiologists and two surgeons blinded to the presence of acetabular fractures in an animal model (a total of 62 porcine hips, 40 with artificial acetabular fractures). Classification of acetabular fractures was adopted from the literature. Interpretation was performed on a workstation using two specific volume rendering algorithms; unshaded and shaded bone. The ROC analysis did not demonstrate any benefit in stereoscopic 3D CT compared with standard 3D CT.


   Introduction
Top
 Abstract
 Introduction
Materials and methods
 Results
 Discussion
 References
 
The management of acetabular fractures requires precise definition of the extent of fracture, comminution and the presence of intraarticular fragments [1]. CT is accepted as an important supplement to conventional radiography in the diagnosis of acetabular fractures [2–4]. Three-dimensional (3D) CT has been shown to influence management, especially in complex cases [5–8].

With the advent of helical CT there has been renewed interest in 3D musculoskeletal imaging. Helical CT has further improved imaging of acetabular fractures by decreasing volume averaging and motion artefacts and has led to improved data sets available for 3D CT [9, 10]. These, coupled with advances in data processing, have led to higher quality 3D CT images [11]. Real-time volume rendering, the most advanced form of 3D CT imaging, is able to display 3D CT images stereoscopically [12, 13]. Although stereoscopic 3D CT has been shown to be useful in the experimental evaluation of acetabular fractures, to our knowledge there has been no study to determine its accuracy and reliability [13–15].

The aim of this study was to compare the discriminatory power of stereoscopic 3D CT and standard 3D CT in the classification of acetabular fractures.


   Materials and methods
Top
 Abstract
 Introduction
 Materials and methods
Results
 Discussion
 References
 
Experimental design
Two experienced radiologists and two experienced orthopaedic surgeons independently analysed a total of 62 porcine hips, 40 with artificial acetabular fractures and 22 normals, using standard and stereoscopic 3D CT and two specific volume rendering algorithms (unshaded and shaded bone). Thus, a total of 992 observations (248 observations per 3D CT technique and volume rendering algorithm) resulted. The hips were obtained from an abattoir and the artificial acetabular fractures were created by an independent surgeon and defined as follows (surgical findings): 4 posterior wall fractures, 8 posterior column fractures, 14 anterior column fractures, 4 transverse fractures, 4 T-shaped fractures, 2 T-shaped plus anterior column fractures, and 4 two-column fractures (Figure 1Go). The classification of acetabular fractures was that of Judet et al [16].



View larger version (94K):
[in this window]
[in a new window]
  		Figure 1.
 
To avoid learning bias, standard 3D CT and stereoscopic 3D CT images were shown to the interpreters in a random order. Standard and stereoscopic 3D CT images were viewed at a workstation (Virtuoso®; Siemens Medical Systems, Forcheim, Germany) equipped with 1280 x 1024 pixel element monitor (Color Graphic Monitor; Silicon Graphics, Mountain View, CA) and diagonal display size of 21 inches. Prior to this, radiologists and surgeons had a training session to become familiar with interpretation of acetabular fractures in pigs. Time constraints were not used.

3D CT technique
Helical CT was performed on a Somatom Plus 4 Power scanner (Siemens Medical Systems, Forchheim, Germany) using a tube potential of 140 kVp, a tube current of 94 mA, 3 mm collimation, pitch of 1.5, reconstruction interval of 2 mm and 0.75 s gantry rotation speed.

CT data sets were transferred via Ethernet network to a 3D Virtuoso® system with free-standing O2 workstation and VA 30 A software (Silicon Graphics, Mountain View, CA). The evaluation of acetabular fractures included reconstruction of a complete CT data set into a volume-rendered 3D format. Two specific volume rendering algorithms were used to classify acetabular fractures as recommended by Kuszyk et al [11]: unshaded and shaded bone. The volume-rendered 3D CT images were reviewed in standard, non-stereoscopic and stereoscopic modes of display with real-time interaction (up to 10 frames per s).

For stereoscopic 3D CT a pair of liquid crystal shutter glasses (Crystal Eyes 2; StereoGraphics, San Rafael, CA) was used. A stereo viewing system like this creates alternating left and right images on a video display screen. The alternating images present the different perspective views of a visual object corresponding to those normally seen by the left and right eye.

The 3D volume sets could be manipulated in standard manner.

Statistical analysis
A differential receiver operating characteristic (ROC) analysis was performed by the two radiologists and the two surgeons blinded to the presence or type of acetabular fractures. The aim was diagnostic accuracy in classification of the artificial acetabular fractures. Readers rated the presence of each acetabular fracture on the basis of a confidence scale of five: 1, definitely not fractured; 2, probably not fractured; 3, questionable; 4, probably fractured; and 5, definitely fractured.

ROC analysis was performed using a specialized computer algorithm (MedCalc Software, Mariakerke, Belgium) created by F Schoonjans. This calculates the area under the curve value (Az) and its standard error using a maximum likelihood estimation technique [17–19]. Factors with Az values greater than 0.80 were considered to have good discriminatory power on the basis of the results of a previous study [19].

Composite ROC curves of the radiological and surgical performance in the two imaging methods (standard 3D CT vs stereoscopic 3D CT) were drawn. Composite ROC curves for each imaging technique were calculated from the average of true-positive fractions achieved by the two radiologists and the two surgeons for each false-positive fraction value [20]. Student's t-test for paired observations with a statistical significance level of p=0.05 was used to test differences in the areas under the curve; the null hypothesis was that composite ROC curves had equal areas beneath them [21].

Sensitivities, specificities, positive predictive values (PPVs) and negative predictive values (NPVs) in the detection of artificial acetabular fractures were calculated for the different 3D CT techniques for radiological and surgical interpreters. Confidence levels of 3–5 were regarded as positive for a fracture, confidence levels of 1–2 were regarded as negative for a fracture. When an acetabular fracture had been incorrectly classified, e.g. T-shaped fracture instead of T-shaped plus anterior column fracture, it was categorized as a false-negative finding.

The interobserver agreement between radiologists and surgeons for lesion detection with each imaging technique was assessed by using the weighted kappa ({kappa}) statistical analysis. In addition, concordance of the imaging technique findings, as determined by radiological and surgical interpreters, with the surgical findings was assessed with the weighted {kappa} statistic. A {kappa} value of less than 0.40 was considered to indicate poor agreement, that of 0.41–0.75, good agreement, and that of above 0.75, excellent agreement [22].


   Results
Top
 Abstract
 Introduction
 Materials and methods
 Results
Discussion
 References
 
ROC analysis
Referring to radiologists, ROC analysis results showed Az values of 0.83±0.04 for standard 3D CT and 0.81±0.04 for stereoscopic 3D CT when using an unshaded bone algorithm (Figure 2aGo). Differences between the two modalities were not statistically significant (p=0.40). Az values were 0.82±0.04 for standard 3D CT and 0.81±0.04 for stereoscopic 3D CT when using a shaded bone algorithm (Figure 2bGo). Differences between the two modalities were not statistically significant (p=0.47).



View larger version (13K):
[in this window]
[in a new window]
  		Figure 2. Composite receiver operating characteristic (ROC) curves of radiologists for standard 3D CT () and stereoscopic 3D CT (---) using (a) unshaded and (b) shaded volume rendering. The composite ROC curves for standard 3D CT are a little bit further to the upper left than are those of stereoscopic 3D CT, but the relative trends among curves are similar.

 
Az values of surgeons were 0.87±0.03 for standard 3D CT and 0.86±0.03 for stereoscopic 3D CT when using an unshaded bone algorithm (Figure 3aGo). Differences between the two modalities were not statistically significant (p=0.42). Az values were 0.82±0.04 for both standard and stereoscopic 3D CT when using a shaded bone algorithm (Figure 3bGo). Differences between the two modalities were not statistically significant (p=0.83).



View larger version (13K):
[in this window]
[in a new window]
  		Figure 3. Composite receiver operating characteristic curves of surgeons for the classification of artificial acetabular fractures with standard 3D CT () and stereoscopic 3D CT (---) using (a) unshaded and (b) shaded volume rendering. The curves were not significantly different.

 
Sensitivity, specificity, PPV and NPV
Using unshaded volume rendering, the overall sensitivity of radiological interpreters was 68% for standard 3D CT vs 70% for stereoscopic 3D CT, while specificity values were 91% vs 86%, respectively. PPVs were 93% and 90%, respectively, and NPVs were 61% for both standard and stereoscopic 3D CT.

Applying shaded volume rendering, the overall sensitivity of radiological interpreters was 68% for standard 3D CT vs 73% for stereoscopic 3D CT, while specificity values were 91% vs 82%, respectively. PPVs were 93% and 88%, respectively, and NPVs were 61% and 62%, respectively.

Using the unshaded bone algorithm, the overall sensitivity of surgical interpreters was 80% for standard 3D CT vs 78% for stereoscopic 3D CT, while specificity values were 96% for both standard and stereoscopic 3D CT. PPVs were 97% for both imaging modalities and NPVs were 72% and 70%, respectively.

Applying the shaded bone algorithm, the overall sensitivity of surgical interpreters was 73% for both standard and stereoscopic 3D CT, while specificity values were 100% and 96%, respectively. PPVs were 100% and 97%, respectively, and NPVs were 67% and 66%, respectively.

Interobserver agreement and interpreter concordance
Interobserver agreement between radiologists and surgeons was good for both standard as well as stereoscopic 3D CT (weighted {kappa}>0.40), regardless of the volume-rendered bone algorithm that had been used.

For radiologists as well as surgeons, weighted {kappa} values for interpreter concordance with surgical findings indicated good agreement for standard and stereoscopic 3D CT ({kappa}>0.40), irrespective of the applied bone algorithm.


   Discussion
Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
References
 
Although conventional radiography is the key to diagnosis and management of acetabular fractures, several authors have emphasized the role of CT, especially in complex cases [1–8, 23–24]. CT has proven valuable in displaying intraarticular bone fragments, joint surface incongruity, femoral head fractures and other unsuspected fractures [2] and shows the size and configuration of associated fracture fragments [1–3, 7–8]. In particular, CT provides information about hip joint stability by revealing comminution of fractures of the posterior wall of the acetabulum [1]. Helical CT has improved evaluation of complex acetabular fractures by decreasing volume averaging and eliminating motion artefacts [9–10, 25–28], so that secondary fracture lines and additional small fragments are now being discovered.

With the continued development of helical CT and recent advances in data collection, computer processing and display technology, there has been renewed interest in 3D musculoskeletal imaging as it is said to be advantageous in complex cases [9–11]. Some investigators have described 3D CT as an excellent method of displaying the complex nature of acetabular fractures because 3D reconstructions accurately present information in the most readily interpretable form [2, 6–7, 9, 25, 28]. These investigators also recommended that 3D CT should be regarded as complementary to axial CT.

The most advanced form of 3D CT, which is currently available in orthopaedic radiology, is volume rendering [10–12]. This is a flexible technique using real-time interaction without the need for preliminary editing [12]. Volume rendering can display an infinite number of imaging planes. Variations in the algorithm may demonstrate both surface and internal detail and allow the interpreter to adjust bone opacity and surface shading to the clinical problem [11, 12]. As a consequence, subcortical lesions, minimally displaced fractures, hidden areas of interest as well as gross 3D relationships may be effectively shown [11]. The option to display 3D CT images in stereoscopic mode has been increasingly used in many orthopaedic applications [13–15]. Several authors have emphasized the potential benefits of stereoscopic display that lie in its ability to separate overlapping fracture fragments and show complex anatomy [13]. However, the usefulness of stereoscopic 3D CT in acetabular fractures must lie in the ability to classify them exactly. For this reason, we compared stereoscopic 3D CT and standard 3D CT in the classification of acetabular fractures in an animal model.

Surgical interpreters showed diagnostic performance of stereoscopic 3D CT similar to that of standard 3D CT in artificially created acetabular fractures regardless of the applied bone algorithm. We showed good interobserver agreement, similar sensitivities and specificities, and similar Az values for the unshaded and shaded volume-rendered bone algorithm.

Although data from radiological interpreters suggested that the sensitivity of stereoscopic 3D CT was slightly superior to standard 3D CT (70% vs 68% for the unshaded bone algorithm and 73% vs 68% for the shaded bone algorithm, respectively), this was likely due to chance. The ROC analysis of radiological interpreters did not demonstrate benefit of stereoscopic 3D CT compared with standard 3D CT.

The relatively low sensitivity values of radiologists compared with surgeons may be caused by limited familiarity with the classification of Judet et al [16].

That Az values in our study ranged from 0.81 to 0.83 for radiologists and from 0.82 to 0.87 for surgeons, showed that, with regard to the degree of difficulty, the distribution of acetabular fractures was within a reasonable range. If we had assessed fractures that were too easy to classify, the Az value would have been close to 1.0. However, this result would have meant a perfect discriminatory power.

The design of our study had certain advantages. First, there was a well defined standard of reference, because acetabular fractures were created artificially. Second, there was no radiation exposure. Third, it permitted a large number of observations so that meaningful qualitative conclusions could be drawn from our ROC analysis. Fourth, standard and stereoscopic 3D CT images were analysed by soft copy using a monitor. In general, viewing of digital images on a monitor display is an advantage because it allows interactive manipulation and exploration of data sets, e.g. window level, center and zoom.

There were various limitations to our study. First, we had to assess juvenile porcine hips in which epiphyseal growth plate fusion might not have occurred. Hence, it was sometimes difficult to determine if an epiphyseal fracture was present or not. These difficulties may be one of the reasons for the relatively low sensitivity. Second, the artificial creation of acetabular fractures sometimes made the shape location atypical; low true-positive values may have resulted. Third, reading bias could not be eliminated because the same data sets were used for the interpretation of standard and stereoscopic 3D CT images, irrespective of the applied bone algorithm.

On the basis of our results, the diagnostic accuracy of stereoscopic 3D CT in the classification of acetabular fractures is comparable to that of standard 3D CT irrespective of the applied bone algorithm, which implies that the additional value of stereoscopic 3D CT is limited in clinical practice.

Received for publication August 29, 2001. Revision received February 6, 2002. Accepted for publication February 14, 2002.


   References
Top
 Abstract
 Introduction
 Materials and methods
 Results
 Discussion
 References
 

  1. Vas WG, Wolverson MK, Sundaram M, Heiberg E, Pilla T, Shields JB, et al. The role of computed tomography in pelvic fractures. J Comput Assist Tomogr 1982;6:796–801.[Medline]
  2. Scott WW Jr, Fishman EK, Magid D. Acetabular fractures: optimal imaging. Radiology 1987;165:537–9.[Abstract/Free Full Text]
  3. Kellam JF, Messer A. Evaluation of the role of coronal and sagittal axial CT scan reconstructions for the imaging of acetabular fractures. Clin Orthop 1994;305:152–9.
  4. Harley JD, Mack LA, Winquist RA. CT of acetabular fractures: comparison with conventional radiography. AJR 1982;138:413–7.[Abstract/Free Full Text]
  5. Burk LD Jr, Mears DC, Kennedy WH, Cooperstein LA, Herbert DL. Three-dimensional computed tomography of acetabular fractures. Radiology 1985;155:183–6.[Abstract/Free Full Text]
  6. Guy RL, Butler-Manuel PA, Holder P, Brueton RN. The role of 3D CT in the assessment of acetabular fractures. Br J Radiol 1992;65:384–9.[Abstract]
  7. Fishman EK, Magid D, Ney DR, et al. Three-dimensional imaging: state of the art. Radiology 1991;181:321–37.[Free Full Text]
  8. Drebin R, Magid D, Robertson DD, Fishman E. Fidelity of three-dimensional CT imaging for detecting fracture gaps. J Comput Assist Tomogr 1989;13:487–9.[Medline]
  9. Fishman EK. Protocols for helical CT of the musculoskeletal system. In: Silverman PM, editor. Helical (spiral) computed tomography. Philadelphia, PA: Lippincott-Raven Publishers, 1998.
  10. Fishman EK. Spiral CT evaluation of the musculoskeletal system. In: Fishman EK, Jeffrey RB Jr, editors. Spiral CT: principles, techniques, and clinical applications. New York, NY: Raven Press, 1995:141–58.
  11. Kuszyk BS, Heath DG, Bliss DF, Fishman EK. Skeletal 3D CT: advantages of volume rendering over surface rendering. Skeletal Radiol 1996;25:207–14.[Medline]
  12. Johnson PT, Heath DG, Bliss DF, Cabral B, Fishman EK. Three-dimensional CT: real-time interactive volume-rendering. AJR 1996;167:581–3.[Free Full Text]
  13. Gautsch TL, Johnson EE, Seeger LL. True three dimensional stereographic display of 3D reconstructed CT scans of the pelvis and acetabulum. Clin Orthop 1994;305:138–51.
  14. Pitman AG, Tress BM, Witte DR. Stereoscopic presentation of computed tomography and magnetic resonance images. Australas Radiol 1999;43:31–6.[Medline]
  15. Haveri M, Suramo I, Junila J, Lahde S, Karhula V. A simple device for the stereoscopic display of 3D CT images. Acta Radiol 1997;38:1053–6.[Medline]
  16. Judet R, Judet J, Letournel E. Fractures of the acetabulum: classification and surgical approaches for open reduction. Preliminary report. J Bone Joint Surg Am 1964;46:1615–46.[Free Full Text]
  17. Metz CE. Basic principles of ROC analysis. Semin Nucl Med 1978;8:283–98.[Medline]
  18. Metz CE. ROC methodology in radiologic imaging. Invest Radiol 1986;21:720–31.[Medline]
  19. Metz CE. Some practical issues of experimental design and data analysis in radiological ROC studies. Invest Radiol 1989;24:234–45.[Medline]
  20. Metz CE. Statistical analysis of ROC data in evaluating diagnostic performance. In: Herbert D, Meyers R, editors. Multiple regression analysis: applications in health sciences. New York, NY: American Institute of Physics, 1986:365–84.
  21. Hanley JA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983;148:839–43.[Abstract/Free Full Text]
  22. Fleiss JL. Statistical methods for rates and proportions (2nd edn). New York, NY: Wiley, 1981:212–36.
  23. Shirkhoda A, Brashear HR, Staab EV. Computed tomography of acetabular fractures. Radiology 1980;134:683–8.[Abstract/Free Full Text]
  24. Gilula LA, Murphy WA, Tailor CC, Patel RB. Computed tomography of the osseous pelvis. Radiology 1979;132:107–14.[Abstract]
  25. Haveri M, Junila J, Suramo I, Lahde S. Multiplanar and 3D CT of acetabular fractures. Acta Radiol 1998;39:257–64.[Medline]
  26. Kreitner KF, Mildenberger P, Rommens PM, Thelen M. Rational diagnostic imaging of pelvic and acetabulum injuries. Rofo Fortschr Geb Rontgenstr Neuen Bildgeb Verfahr 2000;172:5–11.[Medline]
  27. Brooks RA, Ribbans WJ. Diagnosis and imaging studies of traumatic hip dislocations in the adult. Clin Orthop 2000;377:15–23.
  28. Stroszczynski C, Schedel H, Stockle U, Wellmann A, Beier J, Wicht L, et al. Clinical application of multiplanar and 3D reconstruction of spiral CT in diagnosis of acetabulum fractures. Aktuelle Radiol 1996;6:91–5.[Medline]




This Article
Right arrow Abstract Freely available
Right arrow Figures Only
Right arrow Full Text (PDF)
Services
Right arrow Similar articles in this journal
Right arrow Similar articles in PubMed
Right arrow Alert me to new issues of the journal
Right arrow Download to citation manager
Right arrow reprints & permissions
Citing Articles
Right arrow Citing Articles via Google Scholar
Google Scholar
Right arrow Articles by Kickuth, R
Right arrow Articles by Kirchner, J
Right arrow Search for Related Content
PubMed
Right arrow PubMed Citation
Right arrow Articles by Kickuth, R
Right arrow Articles by Kirchner, J

HOME HELP FEEDBACK SUBSCRIPTIONS ARCHIVE SEARCH TABLE OF CONTENTS
BJR DMFR IMAGING  ALL BIR JOURNALS