18 Virtual Endoscopy in Endoscopic Pituitary Surgery
Because of its outstanding imaging quality, endoscopy is gaining increasing acceptance for the transsphenoidal approach to sellar lesions. Due to a steep learning curve, however, special training and preoperative planning are mandatory for the safe application of this technique.
Virtual endoscopy is a concept allowing the user to navigate through computationally reconstructed patient anatomy using a virtual camera, mimicking a real endoscopic approach. It fuses radiologic imaging with advanced three-dimensional (3D) computer graphic techniques to produce views that closely reflect those obtained during physical endoscopy.
Virtual endoscopy can visualize the nasal anatomy and its variations, the location of the sphenoid ostium, the sphenoid sinus septations, and the chambers for improved intraoperative orientation. The pituitary gland, tumor, and adjacent vascular structures can be concurrently demonstrated in relation to the sphenoid sinus landmarks for planning a tailored opening of the sellar floor and tumor removal.
This chapter describes the simulator for transsphenoidal endoscopic pituitary surgery (STEPS), which is based on virtual endoscopy. The system is designed as a tool for training and planning of the endonasal transsphenoidal approach. It has been our experience that virtual endoscopy may increase the safety of the procedure by preoperative visualization of anatomical variations in the individual patient.
At present, the microscopic transsphenoidal approach is considered worldwide the standard neurosurgical technique for the removal of most sellar tumors. Although it has been well established for decades, incomplete tumor resection is not uncommon even for experienced pituitary surgeons.1 Invasive growth with extension into perisellar regions beyond the microscopic field of view may be possible causes of tumor tissue being left behind during surgery.
The use of endoscopes1 ensures excellent illumination due to light emission close to the region of interest2; offers a high-resolution, close-up, and wide-angle view of the anatomy due to the proximity of the lens to the region of interest3; offers the possibility to look “around the corner” by the application of endoscopes at different angles, enabling direct visualization of the removal of tumor remnants in the parasellar compartments4; and offers extended maneuverability of the surgical instruments, as movements are not hindered by a rigid nasal speculum.2 Therefore, the endoscopic technique may increase the extent of tumor removal and the safety of the procedure compared with the standard microscopic transsphenoidal approach.
Although Guiot et al3 made the first attempts to use an endoscope for surgery of sellar pathologies as early as 1963, it was not until recently that a group of neurosurgeons reconsidered and popularized the technique of the purely endoscopic transsphenoidal pituitary approach.4,5 This technique, however, has certain disadvantages, which contribute to its steep learning curve.6 Neurosurgeons are usually well accustomed to the operating microscope but often are unfamiliar with the handling of an endoscope. Furthermore, the endoscope provides a different view of the surgical anatomy due to its two-dimensional image, peripheral barrel distortion, high magnification, and unusual viewing angles.
To overcome these possible risk factors associated with the new technique, teaching and training of the endoscopic transsphenoidal approach are essential for the novice. Endoscopic training, including handling and anatomical viewing skills, is currently provided in workshops at different centers throughout the world. However, cadaver training is costly and often not easily accessible.
Therefore, beginning in 2004, we developed the STEPS prototype based on virtual endoscopy.7 For training purposes, it has since been successfully applied at our workshop facility. In the clinical setting the simulator has proved to be beneficial to the experienced endoscopic surgeon in preoperative planning of surgical cases.8 STEPS has continually been improved by ongoing discussion among trainees, surgeons, and engineers.9,10 This chapter describes our experience with the application of STEPS, its benefits and drawbacks, as well as techniques of visualization and current developments.
Virtual Endoscopy
Virtual endoscopy enables the user to navigate through computationally reconstructed patient anatomy using a virtual camera, mimicking a real endoscopic approach. It fuses radiologic imaging with advanced 3D computer graphic techniques to produce views that closely reflect those obtained during physical endoscopy. Depending on the application, virtual endoscopy can be used either instead of physical endoscopy or as an addition.
David Vining, who was the first to assess computer-generated images of the interior of the colon,11 is often referred to as the inventor of virtual endoscopy. Soon after, various areas of application were presented, including examination of the carotid arteries12 and virtual endoscopy in stereotactic neurosurgery.13 On the basis of this early work, virtual endoscopy became an active area of research, and every year a substantial number of researchers, from both the medical and the technical community, present and assess new techniques and applications. Supported by this wide field of collected knowledge, virtual endoscopy is now making its way from preclinical research into clinical practice.
Virtual colonoscopy was first introduced by Vining11 as a reliable, cost-effective, and less invasive diagnostic alternative to physical colonoscopy. At present, virtual endoscopy is most commonly used in the clinical setting for the detection of colonic polyps. It has proven be an effective technique for finding colonic polyps and has the potential to limit the application of physical colonoscopy specifically to cases in which either a suspicious polyp was found or that were inconclusive in virtual colonoscopy. Another popular area of application is virtual bronchoscopy. It can locate lesions, detect carcinoma, and reliably evaluate airway stenoses.14 Virtual endoscopy has been applied in blood vessels as virtual angioscopy,15,16 for example, in the assessment of abdominal or cerebral aneurysms, carotid stenoses, and atherosclerotic plaques, as well as in the coronary blood vessels, to assess abnormalities of myocardial dynamics.17 Furthermore, virtual endoscopy is currently applied in the small intestines, the stomach, the urinary tract, and the cerebral ventricles.
Some work has been presented on virtual endoscopy in the nose and paranasal sinuses.
Morra et al18 claim that virtual endoscopy provides clear visualization of the anatomical structures of the nasal cavity and sinuses, producing images that are similar to those of conventional endoscopy. Moreover, virtual endoscopy visualizes those paranasal sinuses that are not accessible in conventional endoscopy. The authors state that the main limitations of virtual endoscopy are the arbitrary choice of reconstruction parameters and the scalarization resulting in homogenization of different tissue densities.
Bisdas et al19 claim that in comparison to looking at plain radiographs, virtual endoscopy enables physicians to better understand the complex and varied intracranial structures. They tested virtual endoscopy in 50 patients with the diagnosis of an acute or chronic nasal obstructive disease. The 3D images provided a clear demonstration of the various pathologies within their anatomical contexts. However, this group also experienced difficulties arising from inflexible reconstruction parameters. For example, they point out that their virtual endoscopy software proved inadequate when the nasal cavities were totally filled with soft tissue, such as in extensive acute inflammation or a heavily obstructive disease.
Rogalla20 reported a prospective study in which two radiologists assessed coronal reconstructions and virtual endoscopy with respect to the ease of finding pathologies. In 30 patients, the surgeons were asked to rank the degree of assistance of the preoperative virtual endoscopy. A high degree of similarity between virtual endoscopy and intraoperative aspect was reported.
Due to inflexible surface reconstruction, however, certain anatomical structures (e.g., upper nasal conchae) were not displayed, and the maxillary sinuses could not be approached. Still, virtual endoscopy was found applicable as a preoperative assistance for the surgeon.
Virtual endoscopy for preoperative planning of transsphenoidal pituitary surgery was first presented by Talala et al.21 In addition to standard virtual endoscopy software, nonperspective visualization of the sphenoid anatomy, in which the pre-segmented carotid arteries were displayed, was generated. The authors state that even this nonperspective viewing, which significantly deviates from what is seen during real endoscopic interventions, increases the safety of the procedure.
In recent years, technical advances have led to considerable improvements of processing capabilities of personal computers. This has resulted in enhanced potential for virtual endoscopy software. Motivated by these advancements, we have developed a software package that is not impaired by the aforementioned shortcomings of prior efforts. The STEPS system is described in the following section.
The STEPS System
The STEPS virtual endoscopy system is used for preoperative planning and training of endoscopic pituitary surgery. It is integrated in the commercially available Impax EE PACS system (Agfa-Gevaert Group, Vienna, Austria).
Image Data
Prior to the surgical intervention, standard radiologic images that are used for routine preoperative workup and in traoperative neuronavigation are generated and can be used for virtual endoscopy. Thus, the application of STEPS does not cause additional patient discomfort, radiation hazard, or financial expense.
Due to its high resolution and superior air/mucosa/bone delineation, as compared with magnetic resonance imaging (MRI), a computed tomography (CT) scan is used for reconstruction of the virtual cavities and bony anatomy of the nose and skull base. The CT images used comply with the following protocol: axial plane, 0-zero gantry tilt, 512 × 512 matrix, starting just inferior to the nares and advancing at 1-mm slice thickness.
An MRI is performed for the visualization of anatomical objects of interest. The T1-weighted contrast-enhanced images are used for delineation of the sellar pathology, the pituitary gland, the optochiasmatic structures, and adjacent arachnoid spaces. The proposed sequence at 1.5 tesla is a T1-weighted turbo field echo scan with 20-millisecond repetition time and 4.6-millisecond echo time, one excitation, 25-degree flip angle, 256 × 256 matrix, 2-mm slice thickness. A magnetic resonance angiography (MRA) scan is performed for optimal delineation of the parasellar internal carotid arteries: contrast-enhanced 3D time-of-flight MRA, 512 × 512 matrix, 1.4-mm slice thickness on a 1.5-tesla magnetic resonance (MR) scanner.
For preprocessing of image data, CT and MR scans are stored on the hard disk of the virtual endoscopy workstation.
Software and Workflow (Figs. 18.1 and 18.2)
In addition to the standard virtual endoscopy functionality, the STEPS software package offers the following features:
1. Object processing: Anatomical objects of interest can be segmented by the user and then displayed behind the semitransparent walls of the virtual endoscopy nasal cavity to enhance the information given. Objects of interest are extracted (segmented) from the MR images. Because the visualization of the boundaries of the investigated cavities is based on CT-derived data, the two data sets must be geometrically aligned (registered) to ensure exact correlation of object locations.
Registration is performed automatically by the Impax EE PACS workstation. After registration, objects of interest can be segmented in the MR image and displayed at correct positions during the CT-based virtual fly-through.
Segmentation is mainly performed manually, supported by thresholding techniques. Although time-consuming, manual segmentation is necessary for exact definition of the sellar pathology, pituitary gland, optochiasmatic system, and arachnoid spaces on T1-weighted MRI. Semi-automatic segmentation via threshold settings can be used for the arteries of the circulus arteriosus on MRA. Segmented anatomical structures are stored as objects on the disk.
To display the segmented background objects during the virtual fly-through, each object must be reconstructed from its binary segmentation mask. This is not a trivial problem because the resulting object should have a natural-looking surface while remaining faithful to the segmentation result. In Fig. 18.1A (left), a resulting object that was directly reconstructed from the segmentation mask without any filtering is shown. This is an accurate representation of the segmentation result, but it is of little utility because the sharp edges look quite unnatural. In Fig. 18.1A (middle), the opposite extreme is shown; in this case, the object was filtered without taking care to preserve its features. It is clearly visible that important features are lost, which is inappropriate for a surgical planning application. In Fig. 18.1A (right), the result of applying a reconstruction algorithm we developed for this project is shown; this algorithm preserves all volume elements when filtering the inside-outside classification. Thus, although the resulting object appears acceptably smooth, it still accurately corresponds to the segmentation result.
2. Fast and flexible reconstruction: The virtual endoscopic images are generated using two different reconstruction techniques, one for rendering the standard virtual endoscopy views (the foreground) and one for rendering the background objects. Both techniques are highly optimized in terms of speed and flexibility.
For every frame of the animation sequence, two images, one for the foreground and one for the background, are generated. In a second step, the two images are fused to a single result image. In image fusion, a simple trick is used to improve depth perception. In Fig. 18.1B (left), only the foreground sphenoid sinus anatomy is depicted. The image displayed in Fig. 18.1B (middle) shows the foreground and background objects and was generated using constant foreground transparency. In Fig. 18.1B (right), the transparency was dependent on the distance between the foreground and background, which significantly increases depth perception. Of course, the user can switch to constant transparency at any time to view the whole set of background objects.
The user may also wish to obtain more information about what is behind the surface. STEPS offers a feature that displays surface rigidity, giving information, for example, about whether there is bone directly behind the mucosa. This aids navigation and helps the user to mentally connect the reconstructed image with the real anatomy. Three modes are offered: The fastest is simple color-coding. The second mode is double iso-surfacing, where a second surface is displayed beyond the primary foreground surface, representing firm structures (bone) that are nearby. The slowest, but visually most attractive, method is local direct volume rendering (Fig. 18.3E, middle).
3. Simulation of the surgical procedure: The user can simulate a complete endoscopic intervention, from the nasal to the sellar phase. The process of advancing the endoscope through the narrow nasal cavities and finding the correct way to the sphenoid sinus can be simulated as well as the process of widening the sphenoid ostium, the removal of sphenoid septa, and, one of the most crucial parts of the actual surgical operation, the process of opening the sellar floor. Simulation is used mainly for training purposes but can also provide information about accessibility and the risks and obstacles to expect, which makes it useful also for preoperative planning. For the simulation, it is important to model the degrees of freedom and constraints experienced when working with a rigid endoscope as accurately as possible. Thus the user is confronted with similar challenges as in the real procedure.
During the simulation, the user’s range of view is very limited. It can, as in real endoscopy, be enhanced by switching to an angled endoscope. However, this is often not sufficient to acquire a complete picture of the current state of the simulation, which would be useful especially for training purposes. STEPS provides the possibility of suspending the simulation, in which case the simulation is frozen and the application switches to a temporary free-flight mode. During suspension of the simulation, the endoscope is added to the rendered scene as an additional background object. The user can view the current situation and assess the state of the simulation from arbitrary view points (Fig. 18.3F, right).
The STEPS system provides a virtual surgical instrument that simulates the effect of a bone punch or drill. It can be used to enlarge the sphenoid ostium (Fig. 18.3E, right), to remove septa inside the sphenoid sinus, and to open the sellar floor (Figs. 18.4C, left, and 18.5C