Comparative evaluation of Tuned Aperture ... - BIR Publications

Dentomaxillofacial Radiology (2000) 29, 297 ± 301 ã 2000 Macmillan Publishers Ltd. All rights reserved 0250 ± 832X/00 $15.00 www.nature.com/dmfr

Comparative evaluation of Tuned Aperture Computed Tomography1 for the detection of mandibular fractures MK Nair*,1, RL Webber2 and MP Johnson1 1 Department of Oral and Maxillofacial Radiology, University of Pittsburgh School of Dental Medicine, Pittsburgh, Pennsylvania, USA; 2Department of Radiology, Wake Forest University School of Medicine, Winston-Salem, North Carolina, USA

Objectives: To compare the diagnostic ecacy of Tuned Aperture Computed Tomography (TACT1) with conventional imaging modalities for detection of fractures of the mandible. Methods: Fractures were induced using blunt trauma in human de¯eshed mandibles. Conventional extra-oral and indirect digital images, unprocessed TACT images reconstructed from eight and 16 basis images (BI) and iteratively restored TACT images from eight and 16 BI were used for fracture evaluation. Twelve observers recorded their diagnoses using a ®ve-point con®dence rating scale. The data were analysed using ROC curve analysis. Results: Signi®cant dierences were found (P50.0001) in the areas under the curve (Az): ®lm, 0.6954; digital images 0.6169; TACT unprocessed images using 8 BI, 0.7420; TACT unprocessed images using 8 BI, 0.7667; TACT unprocessed images using 16 BI, 0.7730; TACT iteratively restored images using 16 BI, 0.8143. No observer-based dierences were found. Fractures in the condylar and coronoid regions were more dicult to detect than those in the ramus and body of the mandible. Conclusions: Iteratively restored TACT images generated using 16 BI had a superior diagnostic performance to all the other imaging modalities. Studies are in order to evaluate its in vivo potential. Keywords: tomography, X-ray computed; radiography, panoramic; fractures; mandible

Introduction Diagnosis of fractures of the mandible is primarily eected through the use of physical and conventional radiographic examination. Numerous studies have explored conventional radiography as the modality of choice for the detection of mandibular fractures.1,2 Acquisition of conventional radiographs using projection geometries that are perpendicular to one another helps to delineate the fracture further. Lateral and postero-anterior radiographs continue to serve this purpose and are widely used. Imaging of trauma can be further complicated by the associated clinical condition and lack of cooperation on the part of the patient.3 Repeated imaging procedures with conventional techniques can be time-consuming and uncomfortable for the patient. Therefore, a faster acquisition

*Correspondence to: MK Nair, University of Pittsburgh, School of Dental Medicine, G-120 Salk Hall, 3501 Terrace St., Pittsburgh, PA 15261-1923, USA Received 10 March 2000; accepted 7 June 2000

process without the need to reposition the patient between exposures would be an advantage. Panoramic radiography has been shown to be as ecacious as CT in detecting mandibular fractures although it does not permit examination of the region of interest in threedimensions.2 Occasionally, loss of diagnostic accuracy has been reported while using panoramic images.4 Advanced imaging is usually indicated in instances when there is displacement of the fragments such as in high condylar fractures.2,5 ± 7 CT oers high contrast and the capability of multiplanar and 3-D reformatting.8 The major drawbacks include the radiation burden which increases especially with 3-D CT, artifacts due to metallic and high-density structures in the oral cavity and high cost.1 Additionally, the patient must remain motionless during the entire imaging procedure. The use of 3-D CT has been shown to add little information to that already obtained through physical examination and conventional CT.9 Markowitz et al 2 found that the sensitivity of axial and

Mandibular fractures MK Nair et al

298

coronal CT for angle fractures was only 60% with a high false-positive rate for plain radiographs. Finally, the facility for advanced imaging may not always be available to the oral surgeon to permit multidimensional evaluation. The routine employment of expensive imaging modalities with all of these limitations was the most compelling reason for us to explore the capability of Tuned Aperture Computed Tomography (TACT1), a recently introduced and less expensive option, to detect mandibular fractures. TACT1 is a relatively simple, faster method for reconstructing tomographic images from an arbitrary number of two-dimensional projections acquired with diering projection geometry.10 The technique has been explained in detail in previous studies.10 ± 15 There is a substantial reduction in radiation exposure in comparison with advanced imaging techniques. The three-dimensional image databank permits examination of the anatomy as static slices or in video mode with animation. This study compared six imaging modalities that included conventional ®lm-based radiographs, digital images and various types of reconstructed TACT1 images for their diagnostic ecacy in induced mandibular fractures. The null hypotheses were that for the detection of fractures of the mandible, (1) there was no signi®cant dierence in the diagnostic ecacy of the imaging modalities evaluated, (2) there were no signi®cant dierences between observers for any of the imaging modalities, and (3) there were no signi®cant dierences between reading sessions for any of the imaging modalities. Materials and methods Fifteen de¯eshed cadavers were used in this study. Fractures were induced with blunt horizontal trauma from a steel pipe in approximately half the mandibles at dierent sites in a randomized manner.16,17 This resulted in simple oblique fractures, which were divided into two categories: (1) fractures of the body, angle, ramus, and the parasymphyseal regions and (2) condylar and coronoid fractures. Eight mandibles had fractures (eight mandibles6two sites=16 sites) while the others served as controls (seven mandibles6two regions=14 sites). There were thus a total of 30 sites. A 2-cm thick soft-tissue equivalent material (Radiation Measurements, Middleton, WI, USA) was used to simulate soft tissue scatter. TACT1 images were acquired with a CommCAT imaging unit (Imaging Sciences International, Hat®eld, PA, USA)14,15 using 16 and eight loci in space at angles of disparity ranging from 58 to 108 from an orthogonal projection geometry. The software interface of the CommCAT permits manual programming of the unit using a string command option. The specimens were mounted on a positioning jig to allow placement of

Dentomaxillofacial Radiology

multiple ®duciary markers on the side farthest from the source and close to the receptor. Rigidity of projection geometry is not a pre-requisite for image acquisition. Images were obtained with a conventional Ektavision ®lm-screen combination (Eastman-Kodak, Rochester, NY, USA) using antero-posterior projection geometry. Exposure parameters used were: 65 kVp, 10 mA and 0.7 s with a focal spot ± object distance of 40 cm. The ®lms were digitized using a ¯atbed scanner (DuoScan, Agfa Corporation, NJ, USA) to yield 8-bit images at 800 p.p.i. resolution and saved as uncompressed tagged image format (TIF) ®les. Using the TACT1 workbench, a uniform stack of slices progressing from the anterior to the posterior of the specimen was generated for each of the basis image sets. Iterative restoration (IR) was performed on the images.14,15 A total of four sets of TACT1 images were obtained: TACT1 unprocessed slices, using eight and 16 basis-images (BI) respectively, and TACT1 iteratively restored images using eight and 16 BI. Contrast and brightness enhancements using the TACT1 workbench were done by one of the investigators and saved with the ®les to be loaded for viewing. The other imaging modalities evaluated were: (1) conventional radiographs (panoramic, right and left lateral oblique, and reverse Towne's views), using an Ektavision ®lm-screen system and (2) digitized images using these ®lms. All images were acquired with the CommCAT unit, using the following parameters: (a) panoramic: 68 kVp, 5 mA; (b) lateral oblique: 80 kVp, 10 mA, 600 ms; and (c) reverse Towne's: 80 kVp, 10 mA, 800 ms. All ®lms were processed in an automatic processor (X-OMAT 2000, Eastman Kodak, Rochester, NY, USA), using Readymatic Dental Chemicals (Eastman Kodak, Rochester, NY, USA) at 838F for 5 min. Ground truth was assessed by a direct examination of the specimens. Two investigators blindly recorded the location and course of the fractures on each specimen. Any disruption in continuity of the cortex or any displacement were recorded. Any dierences were resolved by forced consensus. Twelve observers were asked to view all the images and record their diagnoses for the two regions, (i) ramus, angle, body and parasymphysis and (ii) coronoid and condylar, using the ®ve-point confidence scale: (1) (2) (3) (4) (5)

fracture de®nitely absent fracture probably absent uncertain if fracture present or absent fracture probably present fracture de®nitely present

Observers represented a wide cross section of health care professionals and consisted of board certi®ed oral and maxillofacial surgeons, oral and maxillofacial surgery residents, emergency room surgeons and general dental practitioners. All observers had at least 7 years experience in interpreting conventional radio-


graphs. The images were randomly sequenced. All digital images were electronically masked and presented as static slices or in video format. The observers were given the option to enhance the images, but were also informed about any prior enhancement. This was done because a previous study had indicated that observers who are unfamiliar with image processing software, usually employ sub-optimal processing parameters.18 No eort was spared to calibrate the observers. One of the investigators trained them in the use of the TACT workbench, after which they participated in a mock reading session. The observers were then asked to use the 5-point scale on a test sample comprising all types of images, half of which did not have fractures. Consistency in interpretation was established before they were allowed to participate in the de®nitive session. Repeat readings were conducted at least 4 weeks later to evaluate interobserver dierences. Data from the ®rst reading sessions were subjected to the Receiver Operating Characteristic (ROC) curve analysis using ROCFIT software.19 The area under each curve (Az) was determined for all imaging modalities as an indicator of the diagnostic accuracy. Multifactorial repeated measures ANOVA was done to test for signi®cance with respect to the main eects produced by the observers, observer sessions, imaging modality, location of the fractures and any interactions between these eects. Post hoc tests using Tukey's analyses were also done for eects that ANOVA indicated signi®cant. Weighted kappa values were computed to evaluate the level of agreement between the ®rst and second readings.

out to enable pair-wise comparisons by imaging modality. The results are summarized in Table 1. Film-based images were signi®cantly inferior to TACT1 iteratively restored images that were generated using 16 BI (P50.018): the digital images were inferior to all the other techniques. No signi®cant dierences were noticed between the TACT1 modalities. Further analyses were done by modality to test for signi®cance with respect to observers and fracture

299

Results Figure 1 shows the results of the ROC curve analyses. Signi®cant dierences were found for the area under the curve (P50.0001). TACT iteratively restored images using 16 BI had the highest Az (0.8143) while digital images had the lowest (0.6169). ANOVA revealed that two main eects were signi®cant: imaging modality (P50.0001) and observers (P50.0001). Post hoc Tukey's analyses were carried

Figure 1 ROC curves for the detection of all mandibular fractures with the six imaging radiation studied. Key: TACT(8): TACT unprocessed slices generated from a dataset of eight basis images. TACT(8)IR: iteratively restored TACT images generated from eight basis images. TACT(16): TACT unprocessed slices generated from a dataset of 16 basis images. TACT(16)IR: iteratively restored TACT images generated from 16 basis images

Table 1 Matrix of pair-wise comparison probabilities based on imaging modalities: Tukey's HSD (Honestly Significant Difference) Multiple Comparisons Imaging modality

Film

Digital

TACT16

TACT16-IR

TACT8

TACT8-IR

Film Digital TACT16 TACT16-IR TACT8 TACT8- IR

1.000 0.020* 0.930 0.020* 0.600 0.800

1.0000 0.0020* 0.0001* 0.0080* 0.0001*

1.000 0.980 0.790 0.930

1.000 0.930 0.980

1.000 0.800

1.000

* Denotes significance. Key: TACT 8: TACT unprocessed slices generated from a dataset of eight basis images. TACT 8-IR: iteratively restored TACT images generated from eight basis images. TACT 16: TACT unprocessed slices generated from a dataset of 16 basis images. TACT 16IR: iteratively restored TACT images generated from 16 basis images Dentomaxillofacial Radiology


300

location. The results are shown in Tables 2 and 3. There were no signi®cant dierences between the observers in evaluating ®lm-based images for fractures whereas all the other modalities displayed signi®cant observer-based eects. Location of fractures did aect accuracy signi®cantly although fractures in ramus, angle and body of the mandible were easier to detect than those in the condylar and coronoid processes. Interobserver agreement using the overall weighted kappa (Kw) was 0.81. The mean percentage of correct responses (PCR) for each of the category of observers was computed. The ER surgeons had the highest mean PCR of 93.3%, followed by oral surgeons and residents (82%), and general dentists (80%). Discussion The CommCAT was used in this study to acquire the TACT images as it was readily available and the X-ray source is easy to position. Use of a digital sensor would have further improved the signal-to-noise ratio with much lower radiation exposure in comparison with the indirect digital acquisition technique used in this study. The amount of information that is available from TACT1 images and the con®dence that the observers demonstrated in their use point to the fact that it has potential as a superior alternative to conventional techniques in imaging mandibular fractures. A much

Table 2 Probability values from repeated measures ANOVA based on effects produced by observers and fracture location for each imaging modality Imaging modality Film Digital TACT 16 TACT 16-IR TACT 8 TACT 8-IR

All observers

All locations

0.16 0.01* 0.01* 0.01* 0.02* 0.01*

0.01* 0.01* 0.01* 0.01* 0.01* 0.01*

*Denotes signi®cance. For Key see Table 1

Table 3 Post-hoc Tukey's test: pair-wise comparison based on location of fractures of the mandible and observers, within each imaging modality (only statistically significant values indicated) Imaging modality

Location CC vs. AOR

Film Digital TACT 8

P