| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
|
by the American Institute of Ultrasound in Medicine J Ultrasound Med 27:15-24 • 0278-4297 Stereoscopic Evaluation of Fetal Bony StructuresDepartment of Radiology, University of California, San Diego, La Jolla, California USA (T.R.N., D.H.P.); Department of Electrical Engineering, Oregon State University, Corvallis, Oregon USA (M.J.B.); Department of Radiology, Cha General Hospital, Pochon Cha University, Seoul, Korea (E.K.J.); and Department of Radiology, Ulsan University Hospital, University of Ulsan, Ulsan, Korea (J.H.L.). Address correspondence to Thomas R. Nelson, PhD, Department of Radiology, 0610, University of California, San Diego, La Jolla, CA 92037–0610 USA. E-mail: tnelson{at}ucsd.edu
Objective. The purpose of this study was to assess the performance of stereoscopic compared with conventional viewing of 3-dimensional ultrasound (3DUS) data for evaluation of fetal bony structures. Methods. A series of 47 human fetuses were evaluated with conventional 3DUS scanning systems. Twenty-five volumes of the fetal head, thorax, and abdomen were acquired. Volume-rendered images of the fetal cranium and spine were displayed interactively on a real-time stereoscopic graphics workstation. Visualization parameters were interactively optimized. Both conventional and stereoscopic images were evaluated for the clarity of structure visualization (0, nonvisualized; 1, nondiagnostic; 2, adequate; and 3, excellent), the ability to identify key anatomic landmarks (eg, sutures, palate, vertebrae, and ribs), artifacts, and evaluation time. Results. Fetal bony structures, especially high-contrast structures, were readily identified with both conventional and stereoscopic. Overall, stereoscopic viewing provided a statistically significant improvement compared with conventional viewing (P < .01), improved conspicuity of complex bony structures, and added structural detail information that assisted in identification of complex anatomy in 14% of the fetal skull and 26% of the fetal spine cases. Overlapping structures were better identified on the volume-rendered stereoscopic display, with stereoscopic viewing improving differentiation of near and far structures. An interactive display and inclusion of a planar slice review further assisted in identification of structures. The evaluation times were comparable for the two methods. Conclusions. The stereoscopic display of rendered 3DUS data adds valuable information that assists in identification of fetal bony structures, such as cranial sutures and spinal vertebrae, particularly in complex formations. The increasing availability of stereoscopic visualization workstations will offer an additional tool for fetal diagnosis and evaluation.
Key Words: fetal • obstetrics • stereoscopic vision • 3-dimensional ultrasound Abbreviations: LCD, liquid crystal display • MIP, maximum-intensity projection • 3DUS, 3-dimensional ultrasound
Three-dimensional ultrasound (3DUS) imaging is a useful imaging modality that offers improved visualization of fetal anatomy and function.1 Overlying anatomy, shadowing, and complex anatomic shapes can complicate identification of fetal structures. Various display methods are used to enhance visualization, including viewing of multiple slices, volume rendering, and surface rendering.2 The fetal face is often displayed by some type of surface rendering to show the fetal surface.3 Skeletal structures, on the other hand, often are better appreciated with maximum-intensity projection (MIP) methods.4,5
The use of volume data offers the possibility of a detailed review of structures in ways not possible by a planar display. Because most anatomy has some curvature, a volume display can reveal valuable detail regarding complex structures with the proper choice of visualization tools and algorithms. However, even with volume visualization methods, some structures can be difficult to visualize if there are other structures that overlap or obscure each other, even with high-quality images. Interactive adjustment of the rendering algorithm and other visualization parameters such as the angle of viewing, the threshold, a thick slice display, and a multislice display can improve visualization by emphasizing specific anatomic details. Such techniques are essential to producing high-quality images and helping identify the relative location. Interactive volume display methods can assist in identifying the relative locations of structures and can be further enhanced by using the depth cues that arise from relative motion from different depths and shading that emphasizes the relative closeness to the observer by the brightness of the structure.
In some cases, however, it is still difficult to appreciate all of the structural details present either because there are so many or their overlapping overwhelms the viewer. Because volume display methods project a 3-dimensional structure onto a 2-dimensional plane, they are best suited to relatively simple structures. Visualization of more complex structures is challenging at best and requires additional strategies for success. One approach that offers promise is using stereoscopic viewing. By taking advantage of the human visual system, stereoscopic viewing enhances perception of complex structures and complements other volume visualization methods.
Stereoscopic viewing has been shown to improve visualization of spatial relationships. In fact, stereoscopic viewing has a long history,6 dating from the early development of photography that used stereopticons for entertainment7 to geographic surveys using high-altitude imagery8 and more recent uses in popular and scientific arenas,9 including medical imaging.10,11 Regardless of the approach, the underlying concept behind stereoscopic viewing is to provide each eye with an image that views the object of interest from a slightly different viewing perspective (Figure 1
There are various strategies for producing and displaying stereoscopic images.9,12 These include the use of polarizing filters to view separate right and left image pairs, complementary color (red/cyan) image pairs, and liquid crystal light valve glasses that alternately display right and left image pairs to each eye. There also are newer technologies that make use of spinning disks, lenticular screens, and liquid crystal displays (LCDs) that show promise (Figure 2  ).
Combining stereoscopic techniques with interactive visualization including adjustment of rendering parameters and orientation has the potential to improve viewing of internal structures, even in very noisy environments such as ultrasound imaging. The objective of this study was to assess the performance of stereoscopic viewing compared with conventional displays of 3DUS data for evaluation of fetal bony structures.
Fetal Patients In this study, a series of 47 human fetuses undergoing standard fetal sonography were evaluated with conventional 3DUS scanning systems (Voluson 730, GE Healthcare, Milwaukee, WI) with Institutional Review Board approval and after informed consent was obtained. After a routine 2- and 3-dimensional ultrasound scan, 25 volumes were acquired of the fetal head, thorax, and abdomen from different orientations (axial, sagittal, and coronal). Volume data were archived onto a CD-ROM after acquisition and transferred to a graphics workstation for further analysis and review.
Volume Data Review
Volume Data Review Volume data were processed to produce alpha blend and MIP images with depth cuing (Figure 4 ). Volume-rendered images of the fetal cranium, face, spine, and internal structures were displayed on the workstation. Visualization parameters (eg, blend, MIP, and depth) were interactively optimized for standard and stereo techniques.
Stereoscopic Viewing Stereoscopic processing consisted of producing an image pair corresponding to the right and left eye views with a 2.5° angular separation. Reviewers used red/cyan and LCD glasses for the stereoscopic display (Figure 5 ). Evaluation considered the clarity of structure visualization, interactivity, and ability to identify key anatomic landmarks for both standard and stereoscopic visualization methods.
Fetal Skull Analysis Methods We evaluated 20 different landmarks in 102 fetal data sets by conventional and stereoscopic viewing (Table 1 ) for the clarity of structure visualization using an ordinal scale of 0 to 4 (0, non-visualized; 1, nondiagnostic; 2, adequate; and 3, excellent), identification of anatomic landmarks, and artifacts. The total time for evaluating all skull landmarks of the same fetus was recorded. Several days after the conventional viewing review, the complete set of images was shown to the reader again in stereo (via stereo glasses). Visibility scores for each landmark and the total evaluation time were recorded.
Four composite visibility scores were computed for conventionally viewed fetal skull data: an overall sum visibility score for all landmarks and 3 separate sum scores for suture landmarks, face landmarks, and fontanelle landmarks. The same set of 4 composite scores was computed for landmarks viewed in stereo. Paired Wilcoxon rank sum tests were used to compare each corresponding pair (conventional and stereo) of composite visibility scores.
Fetal Spine Analysis Methods
Similarly, 7 composite scores were computed for conventionally viewed and stereo-viewed spine landmarks rated on the same ordinal visibility scale for each reader separately: total composite sum, cervical spine, thoracic spine, lumbar spine, sacrum, thorax, and other. Count and categorical spine variables were not included in the composite scores. Paired Wilcoxon rank sum tests were used to compare each corresponding pair (conventional and stereo) of composite scores. Agreements and disagreements between conventional and stereo viewing were summarized for categorical and count spine data variables.
Evaluation Time
The mean gestational age ± SD of the fetal subjects was 23.6 ± 3.7 weeks. High-contrast fetal skull and spine bony structures were readily visible and identified with both standard viewing and stereoscopic viewing strategies. Overall, the stereoscopic display improved conspicuity of complex bony structures and added structural detail information that assisted in identification of complex anatomy. An interactive display and inclusion of a planar slice review further assisted in identification of structures. Overlapping structures were better identified with the volume rendered stereoscopic display.
Fetal Skull Evaluation
Volume editing with the electronic scalpel was used to remove surrounding tissue from the region of the fetal skull to improve visualization. Editing required 82.8 ± 14.3 seconds with a range of 56.8 to 144.3 seconds depending on the complexity of the overlying structures. Evaluation of the fetal skull volumes after editing showed that conventional and stereoscopic evaluation were comparable in 82% of the cases, with stereoscopic viewing better than conventional viewing in 14% of the cases (Table 4 ).
We also evaluated visualization of symmetric structures in the fetal skull, comparing near and far sutures and fontanelles. Overall, near structures were better seen than far structures (Table 5 ). Stereoscopic viewing assisted in differentiation of fetal skull structures based on their position in the volume, increasing the number of cases in which stereoscopic viewing improved visualization of near structures by 50% compared with conventional viewing.
Evaluation of the relative diagnostic value showed that stereoscopic viewing yielded relatively more studies for which anatomic structures were sufficiently well visualized to make a diagnostic assessment than conventional viewing conditions. However, a key factor was the overall quality of the acquired ultrasound volume. In diagnostically adequate volumes, structures were seen to a greater extent with stereoscopic viewing compared with conventional viewing (Figure 6 ).
Fetal Spine Evaluation Overall stereo viewing provided a higher visibility score than conventional viewing for the fetal spine structures. Both the diagnostic quality of the studies and the number of ossified structures seen were evaluated. Table 6 compares the composite diagnostic quality scores for stereo viewing compared with conventional viewing for all spine data based on the Wilcoxon paired test. Except for the sacrum data, all P values were highly significant for the superiority of stereoscopic viewing over conventional viewing after correction for multiple comparisons. The Spearman correlation coefficient is also given.
Volume editing with the electronic scalpel also was used to remove surrounding tissue from the region of the fetal spine to improve visualization and assist in vertebral body and ossification center identification. Evaluation of edited volumes of the fetal spine and ribs showed that conventional and stereoscopic evaluation were comparable in 67% of the cases, with stereoscopic viewing better than conventional viewing in 26% of the cases (Table 7 ). Evaluation of edited volumes also showed that identification of vertebral bodies and ossification centers was comparable in 55% of the cases, with stereoscopic viewing better in 35% of the cases (Table 8).
Evaluation of the relative diagnostic value showed that stereoscopic viewing yielded relatively more studies for which anatomic structures were sufficiently well visualized to make a diagnostic assessment than conventional viewing. However, a key factor was the overall quality of the acquired volume. In diagnostically adequate volumes, structures were seen to a greater extent with stereoscopic viewing compared with conventional viewing (Figure 7 ).
Evaluation Time Overall, the correlation between conventional and stereo evaluation times was not very high (Table 9 ). Review of the distribution of fetal skull evaluation times showed that stereoscopic viewing times were slightly longer than conventional viewing times (107.4 ± 44.5 versus 99.7 ± 51.6 seconds). However, fetal skull data differences were not statistically significant and showed no systematic differences between conventional and stereo times.
However, review of the distribution of fetal spine evaluation times showed that stereoscopic viewing times were slightly longer than conventional viewing times (235.0 ± 50.5 versus 212.6 ± 60.3 seconds; P < .00001). The more complex fetal spine cases required additional time for identifying small structures that were not as visible with conventional viewing.
Our assessment also compared the performance of an observer fairly experienced in fetal ultrasound with that of an observer who was a general ultrasound imager but was less familiar with reviewing fetal ultrasound volume data. We found that the experienced observer’s times stabilized after approximately 15 cases in the fetal skull, whereas the less experienced observer’s times continued to reduce throughout the evaluation period. The experienced observer also showed a slightly tighter spread of interpretation times compared with the less experienced observer (Figure 8
Stereoscopic displays are known to add information to visualization in many nonmedical as well as medical fields.15 Our understanding of stereopsis continues to increase with more recent contributions from cognitive sciences.16–19 In this study, we have shown that the stereoscopic display of volume-rendered 3DUS data adds useful information that can assist in identification of fetal bony structures such as cranial sutures and spinal vertebrae, particularly in complex formations. However, there is a learning curve for gaining experience in differentiating overlapping structures. The addition of visual depth cues can enhance visualization of structures but also can contribute complexity to the visual scene, complicating and extending interpretation times. Furthermore, some individuals may have a more difficult time using stereopsis than others, suggesting that in addition to learning, there may be fundamental limitations on some individuals regarding their ability to use stereopsis.20,21 These effects are noted to increase with age.22
In our study, stereoscopic viewing improved visualization of small structures when there were multiple overlapping structures. Identification of high-contrast structures such as the fetal skull or spine showed improvements with stereoscopic viewing compared with conventional viewing. We also noted that stereoscopic viewing enhanced visualization of structures near the observer compared with those that were far from the observer (Table 5 Overall, whereas stereoscopic viewing improved visualization of small structures in a complex view field, it also complicated interpretation by providing additional information that required additional interpretation time compared with conventional viewing. Fetal skull data composite visibility scores indicated a statistically significant and consistent improvement in visibility with stereo viewing. Spine data composite visibility scores also indicated a statistically significant and consistent improvement in visibility with stereo viewing. Spine data review counting structures visualized yielded more structures with stereo than conventional viewing. Skull data review times were not statistically significantly different for either conventional or stereo viewing. Spine data required more time to evaluate with stereo than conventional viewing. Overall the correlation between time spent with conventional and stereo viewing was not very high. Standard viewing without motion cues made it difficult to identify many structural relationships, whereas stereoscopic viewing without motion was better able to provide cues regarding structural relationships. Overall, the best performance was obtained by combining motion cues with stereoscopic viewing.23 Interactive viewing that includes modification of both rendering parameters and the viewing angle is an essential feature of any volume visualization system. Interactive review of the volume data is probably the most important factor in structure identification for both conventional and stereoscopic viewing. Relative motion (ie, rotation of the volume) of cranial structures provided valuable cues regarding locations, further enhancing stereoscopic viewing in addition to static views. Editing of volumes by removing unwanted (overlying) echoes was also important in optimizing viewing. Fortunately, recent advances in computer hardware and software performance have made such stereoscopy affordable on desktop computers, facilitating viewing of volume data in offices as well as ultrasound equipment and laboratories. More powerful laptop computers make it possible to interactively display volume data in educational and scientific meetings, enhancing comprehension of anatomic structures. Further advances in real-time 3DUS imaging will provide new opportunities for guiding minimally invasive interventional procedures.24 Stereoscopic viewing is not widely available for medical displays at this time. However, a red/cyan display can be accomplished on most displays with little additional hardware or computational requirements. Although the red/cyan display is quite satisfactory for many displays and provided comparable stereoscopic effects as light valve LCD glasses in this study, the overall feel is less satisfying than with other approaches because of the false color nature of the process. Alternatively, light valve LCD glasses remain relatively expensive and require additional hardware for 3-dimensional visualization, which limits their use at the present time. Other types of display technology potentially will continue to make it easier for stereoscopic displays to enter general use. Finally, the increasing availability of stereoscopic visualization workstations offers an additional tool for fetal diagnosis and evaluation. Our results suggest that this tool can provide additional information regarding fetal structures, particularly in complex anatomy, and thus has a role in fetal imaging and diagnosis in the future.
We thank Tanya Wolfson (Biostatistics and Bioinformatics Laboratory, University of California, San Diego) for assistance with the statistical analysis. This work was supported in part by an equipment loan from GE Healthcare (Milwaukee, WI). Presented in part at the 13th World Congress on Ultrasound in Obstetrics and Gynecology, Paris, France, August 31–September 4, 2003; and the 2004 AIUM Annual Convention, Phoenix, Arizona, June 20–22, 2004. Received June 26, 2007, from the Department of Radiology, University of California, San Diego, La Jolla, California USA (T.R.N., D.H.P.); Department of Electrical Engineering, Oregon State University, Corvallis, Oregon USA (M.J.B.); Department of Radiology, Cha General Hospital, Pochon Cha University, Seoul, Korea (E.K.J.); and Department of Radiology, Ulsan University Hospital, University of Ulsan, Ulsan, Korea (J.H.L.). Revision requested June 28, 2007. Revised manuscript accepted for publication August 20, 2007.
|
|||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
Top
Abstract
Introduction
