| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
Departments of Internal Medicine, Radiology, and Biomedical Engineering, University of Iowa, Iowa City, Iowa
Correspondence and requests for reprints should be addressed to Geoffrey McLennan, M.D., Ph.D., Departments of Internal Medicine, Radiology, and Biomedical Engineering, University of Iowa, Iowa City, IA 52242. E-mail: geoffrey-mclennan{at}uiowa.edu
ABSTRACT
Virtual bronchoscopy is rapidly providing a software solution for visualization and measurement of the human airway tree with data derived from multirow detector X-ray CT scans as the most common data source. Increasingly, other imaging data sources to create three-dimensional image renderings of the bronchial tree are being used including magnetic resonance imaging, ultrasound, and the digital color image taken by the bronchoscope itself. Software solutions now exist for providing simple renderings of the bronchial tree through which a fly-through of the airway lumen by the center line of the airway can be added (the fly-through mimics the view that a real flexible bronchoscope affords the operator). The images so produced are visually accurate (i.e., they appear very realistic) and with good software now available are also analytically correct (i.e., measurements taken from these images are accurate). More advanced virtual bronchoscopic applications, including image-based pathfinding to mediastinal and peripheral lung structures, are also in development. Synergistic datasets composed of image data from multiple image sources are also being developed.
Key Words: bronchoscopy • CT scan • imaging
Virtual bronchoscopy is the descriptive term given to representations of the bronchial tree and surrounding structures created from spatial information derived from imaging sources other than the bronchoscope itself. Initially applied to the two-dimensional and later three-dimensional representations of the bronchial tree obtained from X-ray computed tomography (CT) (1), the term equally applies to similar images derived from magnetic resonance imaging (MRI), from ultrasound imaging, and from image-processing techniques applied to the standard two-dimensional modern bronchoscope image (2).
X-ray CT of the lung produces two-dimensional images (a cross-section of the thorax at the slice point) with the minimal x,y resolution in this image referred to as a "pixel" and the depth of the slice adding a z direction to that pixel; this volumetric minimal resolution image is referred to as a "voxel." Over the last 10 to 15 yr, the x,y and z resolutions of the CT scans have dramatically improved along with improved scan acquisition times. Current CT scanners (64 multirow detector CT devices [MDCT]) now produce x,y and z resolutions of the order of 0.6 mm. If these two-dimensional x,y slices are stacked one on top of the other, maintaining their alignment, then it becomes very clear that a high-resolution three-dimensional image of the thorax can be obtained. This three-dimensional image contains all of the structures within the thorax, including the airways (where there is natural contrast between tissue and air), the mediastinal blood vessels (where, with contrast injection, there is discrimination between blood vessels and other soft tissue structures), and the mediastinal lymph nodes (where there is often significant contrast differences between the lymph node and surrounding mediastinal fat). The three-dimensional image of the thorax obtained through MDCT is a digital image, with each voxel having a defined spatial dimension and gray-scale characteristics. As with any digital dataset, the three-dimensional image of the thorax can be analyzed, digitally stored and retrieved, and displayed as an image-based informational structure. Similar structures within the three-dimensional image dataset can be digitally removed from the remainder of the image for later visualization and analysis.
Using these principles of digital technology it is reasonable to expect that the bronchial tree (and the surrounding structures if needed) can be identified, then removed from the larger image structure and evaluated in three-dimensional digital space. This is the basis of CT-based virtual bronchoscopy and similar to the virtual bronchoscopy views derived from other digital image sources. One issue is that the scanner hardware technology has exploded at such a rate that clinical software solutions have not kept up. This is now starting to change (3).
SIMPLE CT-BASED VIRTUAL BRONCHOSCOPY
In 1996 the author performed his first clinical CT-derived virtual bronchoscopic images. The patient was a 74-yr-old female smoker who presented with significant dyspnea and who could not lie flat. Chest radiograph was unremarkable. Bronchoscopy performed with her sitting showed a large intrabronchial tumor obstructing the distal trachea, with an airway anterior to this reduced to a slit. The question in this high-risk patient was what was the distal extent of the tumor (e.g., did it involve the left and right main bronchi) and what was the length of the obstructing segment. Finally, there was interest to know the relationship of the aorta to the mass lesion, given that laser resection through the rigid bronchoscope was planned. With systemic corticosteroids relieving dyspnea somewhat, a volumetric CT scan was obtained the following day, and the major airways and the aortic arch segmented. Airway segmentation is the identification of the airways on each CT slice, with their walls outlined so that the computer can recognize the outlined region as airway; the removal of these identified and outlined structures from the rest of the image; and the stacking together (or interpolation) of these common images as an airway tree. The same process can be followed for any other structure. The segmentation was performed manually slice by slice, taking several hours each for the airways and the aorta. These results are shown at http://dpi.radiology.uiowa.edu/nlm/app/lngtumor/trcharch/case1.html. Subsequent to the laser procedure, the patient had radiation treatment and developed a bronchoesophageal fistula as can be seen in the November 1996 images on the website. This was managed with stenting and the patient did well for another 12 mo. The procedure, which can now be preplanned together with the anesthetist, went well.
Since this time, significant effort and resources have been put toward automatically segmenting the airway tree from the CT-obtained images in academic research laboratories throughout the world, and in industry. Most of the major CT manufacturers now offer virtual bronchoscopy software with their machines, and there are several smaller companies offering virtual bronchoscopy software. For software packages to be useful, three conditions must be satisfied. First, they must provide adequate visualization of the three-dimensional airway structures and accurately depict the relationships of the bronchial tree with surrounding structures, such as major blood vessels. Second, the visualized results must be not only anatomically correct but provide accurate measurements of the airway lumen for length and diameter of suspect segments. This is extremely important for interventional pulmonology applications, such as bronchial stent placement. Finally, the software must be available in a usable format at the point of service for the medical care or procedure being performed. In the author's situation, he analyzes and visualizes the virtual bronchoscopy information in the bronchoscopic laboratory before and during procedures. This is a very important aspect of the clinical acceptance and use of this new technology.
In the first 50 cases to which the author has applied this technology, and subsequently the several hundred that he has performed over the last 10 yr, the depiction of the bronchial tree from CT scan data has been found to be very useful in the following circumstances. It is useful for demonstrating focal stenoses within the tracheal bronchial tree that might otherwise be missed by an observer reading only the two-dimensional CT images. Missed lesions are particularly likely in the upper trachea and in the right main bronchus where a stenotic segment might only appear on one two-dimensional slice but with visualization through the three-dimensional display, the stenotic segment is easily seen. The author has used the virtual bronchoscopy images for the evaluation of the bronchial tree distal to a known stenosis. The accuracy of virtual bronchoscopy techniques with real bronchoscopy findings is high (4, 5), and this will improve further as CT scanning protocols improve. This is particularly useful for those conditions where there may be multiple stenoses, such as Wegener's granulomatosis (6), or secondary cancers where the traditional bronchoscope cannot visualize past the first obstructing segment. The author has also used virtual bronchoscopy techniques in the follow-up of patients with known bronchial stenosis, such as right main bronchus stenosis at the anastomosis site after lung transplantation where, after an initial procedure, such as balloon dilatation, the stenosis can be followed sequentially using virtual bronchoscopic techniques without putting the patient through a further real bronchoscopic evaluation. Furthermore, the author has found that virtual bronchoscopy techniques provide useful information about the relationship of any bronchial abnormality of surrounding structures, such as the manubrium sterni in a high tracheal stenosis where there may be a question as to whether tracheostomy can be safely performed (Figure 1 [p. 504]). Finally, the author uses these virtual bronchoscopic procedures to help with bronchial stent sizing before the procedure where one can plan both the length and diameter of the stent, or for an initial assessment of balloon sizes for balloon dilation within the airway (7). This preplanning ensures that the required stent is in the inventory and that the inserted stent fits the airway appropriately, leading to less stent migration and less granulation tissue at the ends of the stent. It seems clear that the virtual bronchoscopic view of the airway, if both anatomically accurate and providing accurate measurements, is an important new addition in the pulmonary physician's practice, and is especially useful in evaluating the airway before initial or sequential real bronchoscopies where an interventional procedure might be considered. These images reduce the chances of significant surprises occurring during any procedure, and allow the patient and family to be as fully informed as the physician of record.
These descriptions are that of simple virtual bronchoscopy techniques and applications. This forms the current standard practice for interventional pulmonology procedures where preplanning of complex procedures is essential in improving operative and postoperative outcomes. For review of many of the author's procedural cases, readers are directed to the following website, which lists many examples of the application of virtual bronchoscopy in interventional pulmonology practice: http://dpi.radiology.uiowa.edu/nlm/app/atlas/welcome2.html.
COMPLEX VIRTUAL BRONCHOSCOPY
Complex virtual bronchoscopy procedures are now moving from the research laboratory into clinical studies and some of them will be in clinical practice over the next 1 or 2 yr.
The first of these complex virtual bronchoscopy procedures examined here uses the data contained within the complex three-dimensional image for procedure guidance within the mediastinum and hilar structures (8, 9). A simple example is understanding where a mediastinal lymph node is in relationship to the bronchial tree. In traditional bronchoscopy, only the airway lumen is visualized, although the bronchoscopist knows that somewhere on the other side of that nontransparent bronchial epithelium is the lymph node that is the target for a transbronchial needle aspiration or core biopsy. Using the virtual bronchoscopy information obtained from the three-dimensional MDCT dataset, a virtual bronchoscope (placing a visualization tool where the real bronchoscope might be within the lumen) can show precisely the same image as the real bronchoscope does, with gray-scale rendering rather than the color rendering from the real bronchoscope image. With these two images side by side (the virtual bronchoscopic visualization and the real bronchoscope visualization of the same region), the next step is to make the virtual bronchoscope bronchial wall transparent. In this manner, as with manipulation of many digital images in this current era, the structures through the bronchial wall can now be visualized. Using the extraordinary power of the human brain for image processing, or an associated computer script, the lymph node in question, now visible on the virtual bronchoscope images, can be transferred to overlay the real bronchoscopic images. The bronchoscopy operator can, with a great degree of confidence, know where the target lymph node is (Figure 2 [p. 505]). This process can be simplified before the real bronchoscopic procedure in a preplanning step where the target region, be it lymph node or some other structure, can be rendered as a region of interest and displayed as a color object in both the virtual bronchoscopic and later the real bronchoscopic images. There are several computer programs that have been developed for this application that are currently in phase III clinical studies, with early results suggesting satisfactory ease of use and improvement in the biopsy return. These phase III clinical studies have used experienced bronchoscopists as the operating gold standard, and one imagines that the average bronchoscopist will have significantly better yields with transbronchial needle procedures using this type of image-guided computer assistance, again at the point of service in the bronchoscopic laboratory. Early results suggest a greater than 90% success rate for mediastinal and hilar lymph node biopsies. Clearly, these early successes open the possibility of sampling mediastinal structures precisely and reliably. This also opens up the possibility of localized application of nonspecific or specific therapies.
A second advanced virtual bronchoscopic application is that of pathfinding to a peripheral region of interest within the lung (10). Given the respiratory motion that occurs during breathing, transcutaneous approaches to the moving lung may not be satisfactory. By approaching the peripheral lung through the bronchial tree, movement is not an issue because the parenchymal lung lesions move synchronously with the airway and with any device that is within the airway. With the significant improvement in MDCT scanners, seven or eight generations of airways can now be automatically extracted and evaluated. The simple application here is that if the trachea is the beginning point and if a pulmonary parenchymal abnormality (pulmonary nodule or region of pulmonary emphysema) is the targeted end point, then appropriate software can interrogate the three-dimensional image dataset and provide a pathway through the airway to the lesion. This pathway can then be followed by the bronchoscopist during a real bronchoscopy procedure and the correct airway pathway to the lesion quickly cannulated using a Teflon-coated tube that is very similar to the internal coating of the standard bronchoscopic instrument channel. Once the Teflon access tube is in place, then multiple probes can be placed either to brush or biopsy or optically or by ultrasound sample the lesion of interest. Ultrathin bronchoscopes can be used in a similar manner (11). Using these sorts of approaches in early studies, 80% of peripheral lung lesions can be easily and satisfactorily sampled. These pulmonary pathfinding applications are in clinical studies and are being developed by a number of companies for medical application. They may have a synergistic role also when coupled with magnetic tracking devices.
The third advanced virtual bronchoscopy application involves the targeting of the peripheral lung for endobronchial valve-type procedures in the management of pulmonary emphysema, so-called endobronchial lung volume reduction surgery (12). Here, the information that is required is the state of the lung parenchyma and the extent of emphysema in each segmental region together with the anatomic configuration and size of the subtending airway segments. There are several companies now with hardware solutions for the endobronchial management of emphysema and most of them are in human clinical studies. To shorten procedure times, to improve accuracy of device placement, to reduce medical error, and to educate the patient and families, application of advanced virtual bronchoscopy techniques is an obvious solution. For instance, if the right upper lobe is being targeted (in some studies, this is through human read of the CT scan; in others, it is by computer read) it is unclear to the operating interventional pulmonologist how many segments might require a device to be placed and whether the segment lengths are adequate for the valve placement. Such planning including valve sizing should be performed where possible before the procedure by the management team. In the immediate future, rather than targeting lobar airways with therapeutic devices, it is likely that segmental airway devices will be placed in those airways that subtend areas of severe emphysema. In the lobar-only approach, regions of the lobe that may not have much emphysema are treated unnecessarily. Segmental airway targeting will certainly require computer assistance and procedure planning to improve accuracy of device placement. An example of a software solution for valve and other endobronchial device placement for the management of emphysema and of pulmonary pathfinding is shown in Figure 3 (p. 505). This software also automatically labels the segmented airways to reduce confusion (13).
At the beginning of this article, it was indicated that virtual bronchoscopy images apply not just to CT-derived data, but also to data from MRI, from ultrasound, and from the real bronchoscopic images themselves. MRI- and ultrasound-derived data are not discussed further except to make clear that there is substantial work being performed in both of these fields as they apply to virtual bronchoscopic applications. What is discussed in the remainder of this article are the virtual bronchoscopic images obtained from regular two-dimensional bronchoscopic images, together with the suggestion that fusion of the multiple image datasets relating to the bronchial tree (CT, ultrasound, and real bronchoscopic) is likely to be synergistic in regard to providing important detail about the state of a subject's airways particularly for early disease, such as lung cancer in the central airways.
The bronchoscopic images obtained through a real bronchoscopic examination are now also digital images. As such, there is an extraordinary amount of data. As with digital radiology–type studies, much of the data generated are not used by the clinical examiner. The first digital data in modern bronchoscopy relate to the color of the airway mucosa. Here in the bronchoscopy laboratory one often refers to "redder than normal" or "pale mucosa" without any analytic understanding as to what that means and without any studies to indicate the intraobserver or interobserver variability. This color information can now, through computer means, be defined and recorded in a pixel-by-pixel basis and cross-compared with other normative data or with disease states (14). This advance has only been made possible because of the digital nature of the bronchoscopes. It does allow for comparative studies on airway mucosal color to be undertaken for the first time and for regions of bronchial mucosa that lie outside of the normal range to be highlighted for special consideration by the bronchoscopist, such as possible biopsy or, in the future, some form of optical sampling. Because color assessment is so important in bronchoscopic practice (e.g., this defines inflammation, and cancer) it seems reasonable that the future will see some form of analytic color bronchoscopy as a component in the clinical workspace.
Because the bronchoscopic color information is digital, this can be related to other digital imaging modalities, such as CT. Indeed, the color mucosal information from a regular bronchoscopy can be painted on the virtual CT bronchoscopic images with a view to highlighting structures that might be topographically abnormal (i.e., for shape) and abnormal for color (15). Computer software for these fused applications has also been developed and is undergoing further clinical evaluation.
The two-dimensional bronchoscopic image also contains shadows attributed to the light source and its distribution, both of which are known. Taking account the shadowing through a method known as "shape from shading" (16), a three-dimensional bronchoscopic image can be computer rendered from the two-dimensional images that most bronchoscopists are used to viewing. This type of virtual bronchoscopic image has only been developed for segmental use within the airway with a view to augmenting structural abnormalities that might not be immediately obvious to the bronchoscopist working only in two-dimensional space. These images can be fused with the MDCT dataset to provide synergistic information. A structural segment of the bronchial tree can now be defined combining a three-dimensional MDCT scan virtual bronchoscope image, with the digital color information from real bronchoscopy, with the shape from shading three-dimensional real bronchoscopic virtual bronchoscope images, providing a composite structural virtual bronchoscopic image. If a functional image is fused, such as fluorescein angiography taken at fluorescent bronchoscopy, then for the first time one can evaluate the segmental airway with multiple digital datasets providing a four-dimensional virtual bronchoscopic evaluation (17). An example of this is shown in Figure 4 (p. 505). The three-dimensional structural information combined with airway mucosal changes over time from real-time bronchoscopy provide the four-dimensional virtual bronchoscopy. This has now been achieved; however, clinical use needs to be defined (17).
CONCLUSIONS
Virtual bronchoscopy initially referred to the simple depiction of the bronchial tree as a three-dimensional image usually from CT data. This simple approach is by itself not enough for useful clinical application; however, if combined with accurate measurements in addition to the visualization, then virtual bronchoscopy of this type has a significant contribution to pulmonary practice particularly for interventional pulmonology. This technology must be available at the point of service. Advanced virtual bronchoscopy applications are now in clinical trials including pulmonary pathfinding for peripheral lung lesions and mediastinal biopsies, and preplanning for airway procedures increasingly associated with emphysema management. Four-dimensional virtual bronchoscopy has now been achieved in the laboratory situation and might also shortly enter clinical studies.
By using the patient's own internal structure and function as assessed by these virtual bronchoscopic applications for the preplanning of complex diagnostic or therapeutic procedures, a boundary has been crossed. This personalization of health care delivery represents a fundamentally important shift, complementing nicely similar shifts in genomic research, and seems important for improving the outcome of health care.
FOOTNOTES
The color figures for this article are on pp. 504–505.
Conflict of Interest Statement: J.S.F. does not have a financial relationship with a commercial entity that has an interest in the subject of this manuscript. G.M. is part owner of VIDA Diagnostics and Endographics. He has through the University of Iowa applied for a patent for the lung parenchymal analysis package. This has been approved. Several other patents are pending.
(Received in original form August 1, 2005; accepted in final form September 27, 2005)
REFERENCES
Related articles in Proceedings of the American Thoracic Society:
This article has been cited by other articles:
|
|
 |
|
  C. S. Rogers, W. M. Abraham, K. A. Brogden, J. F. Engelhardt, J. T. Fisher, P. B. McCray Jr., G. McLennan, D. K. Meyerholz, E. Namati, L. S. Ostedgaard, et al. The porcine lung as a potential model for cystic fibrosis Am J Physiol Lung Cell Mol Physiol, August 1, 2008; 295(2): L240 - L263. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 W. A. Ames, D. B. Macleod, A. K. Ross, J. Marcus, and S. Mukundan Jr The Novel Use of Computer-Generated Virtual Imaging to Assess the Difficult Pediatric Airway Anesth. Analg., May 1, 2007; 104(5): 1154 - 1156. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
