An Image-Guided Planning System for Endosseous Oral ... - IEEE Xplore

842

IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 17, NO. 5, OCTOBER 1998

An Image-Guided Planning System for Endosseous Oral Implants Kris Verstreken,* Johan Van Cleynenbreugel, Kirsten Martens, Guy Marchal, Daniel van Steenberghe, and Paul Suetens, Member, IEEE

Abstract— A preoperative planning system for oral implant surgery was developed which takes as input computed tomographies (CT’s) of the jaws. Two-dimensional (2-D) reslices of these axial CT slices orthogonal to a curve following the jaw arch are computed and shown together with three-dimensional (3-D) surface rendered models of the bone and computer-aided design (CAD)-like implant models. A technique is developed for scanning and visualizing an eventual existing removable prosthesis together with the bone structures. Evaluation of the planning done with the system shows a difference between 2-D and 3-D planning methods. Validation studies measure the benefits of the 3-D approach by comparing plans made in 2-D mode only with those further adjusted using the full 3-D visualization capabilities of the system. The benefits of a 3-D approach are then evident where a prosthesis is involved in the planning. For the majority of the patients, clinically important adjustments and optimizations to the 2-D plans are made once the 3-D visualization is enabled, effectively resulting in a better plan. The alterations are related to bone quality and quantity (p < 0:05), biomechanics (p < 0:005), and esthetics (p < 0:005), and are so obvious that the 3-D plan stands out clearly (p < 0:005): The improvements often avoid complications such as mandibular nerve damage, sinus perforations, fenestrations, or dehiscences. Index Terms— Dental implants, image-guided therapy, OpenGL, OpenInventor (OI), surgical planning.

I. INTRODUCTION

O

RAL endosseous implants are used to rehabilitate (partially) edentulous patients. The Br˚anemark system is one of the clinically best documented implant systems [1], [2]. In this system, a fixed prosthesis is mounted on titanium cylinders that are inserted into predrilled holes in the jaw. Through the use of pure titanium in combination with an atraumatic surgical technique an intimate bone apposition to the implants is obtained and a very high long-term cumulative success rate ( 95%) can be reached. A patient who is considered for receiving endosseous oral implants undergoes a thorough clinical and radiological examination before being elected for surgery [3]. The examination Manuscript received March 30, 1998; revised July 2, 1998. The Associate Editor responsible for coordinating the review of this paper and recommending its publication was M. Viergever. Asterisk indicates corresponding author. *K. Verstreken is with the Laboratory for Medical Image Computing, Katholieke Universiteit Leuven, ESAT/Radiology, Univerity Hospital Gasthuisberg, Herestraat 49, 3000 Leuven, Belgium. J. Van Cleynenbreugel, K. Martens, G. Marchal, and P. Suetens are with the Laboratory for Medical Image Computing, Katholieke Universiteit Leuven, ESAT/Radiology, Univerity Hospital Gasthuisberg, 3000 Leuven, Belgium. D. van Steenberghe is with the Department of Periodontology, Faculty of Medicine, Katholieke Universiteit Leuven, 3000 Leuven, Belgium. Publisher Item Identifier S 0278-0062(98)09100-9.

can consist of a radiographic status (a number of small intraoral radiographs), an orthopantomogram (both X-ray source and film rotate around the patient’s head) or ortho-radial tomograms. For clinically more difficult cases where the highest accuracy is needed a computed tomography (CT) scan is still the method of choice [4] and does not necessarily result in a higher dose [5]. We will further focus on this method since it is also for these more difficult cases that a planning is most useful. The acquired CT images are used during surgery, where necessary in combination with available radiographs. Combining all these data with his clinical experience, the surgeon reflects the mucoperiosteum (the gums), inspects the bone structures, drills the planned number of holes, and then inserts the implants. Angulation and depth are taken from measurements on the images, but eventually adjusted on the basis of intraoperative information, such as the tactile sensation of the bone density experienced during drilling. After insertion of 4–6 implants for one jaw, the mucoperiosteum is closed. A healing period of several months is respected before the gums are reopened, this time above the implants, and the top connectors (abutments) are placed whereupon the fixed prosthesis will be mounted. This superstructure is designed to fit the existing implant situation. In full edentulism 4–7 implants can carry a prosthesis with a full dental arch. However, a good surgical technique alone is not sufficient, several important design criteria have to be met for the procedure to be a success. First, it is beneficial for a good osseointegration that the implant is inserted in well-mineralized bone of sufficient height, and that a (bi)cortical contact is obtained. Second, the axes of the implants should be inclined toward the occlusal force vectors to better withstand the relatively high chewing forces. And third, the implants should be placed in such a manner that the prosthetic superstructure can mask them so as not to interfere with oro-facial aesthetics. Aside from the design criteria, there exist constraints inherent to the anatomy of the region. Vulnerable structures such as the mandibular nerve and, to a lesser extent, the sinus cavities, must be avoided. These constraints often severely limit the number of available locations and so result in an absolute contraindication for endosseous oral implants for a significant number of patients. They are the reason that a preoperative plan is routinely made for the more demanding cases. A preoperative planning aims at optimizing all the design specifications, which can related to bone quality/quantity, biomechanics and esthetics, while remaining within the constraints.

0278–0062/98$10.00  1998 IEEE

VERSTREKEN et al.: IMAGE-GUIDED PLANNING SYSTEM FOR ENDOSSEOUS ORAL IMPLANTS

The manual planning procedure previously in use at our center is described first: There exist commercially available software packages (e.g., Dental CT, Siemens, Erlangen, Germany) that perform on the scanner console a reformatting specifically tailored to oral applications. The resulting set of images consists of axial slices, an overview axial slice with the curve, panoramic images along the curve, and reslices orthogonal to it. The reformatting process is guided by a technician on the CT console: On one of the axial slices, a flexible curve is aligned with the centerline of the jaw arch, and multiplanar reformatted images are constructed orthogonally upon this curve. These images are printed to film and are then used to determine whether implant surgery is possible or not and to define a case-specific success percentage related to the individual bone parameters of the patient. Implant lengths are derived from jaw bone height and width, measured with a simple transparent ruler overlaid on the film. Positions are given numerical values related to standard tooth locations. The information of this planning is recorded in the patient file and used as such during surgery. There also exists planning software that is based on the axially reconstructed images from the CT scanner, such as the software mentioned in [6] and the commercial package Sim/Plant (Columbia Scientific, Inc., Columbia, MD). These packages are inherently two-dimensional (2-D) and can be compared to a computerized version of the manual planning method. However, several specifications and constraints can only be evaluated using a three-dimensional (3-D) approach due to their spatial nature. Specifications arising from biomechanics or esthetics are often difficult to incorporate into a 2-D planning method. Biomechanics involves the orientation of the implant axis toward the occlusal plane of the teeth and this requires alterations in all directions of the axes. Esthetics requires a frontal view of the implant-superstructure relationship together with the teeth of the prosthesis. Planning packages that allow an interactive 3-D visualization do exist, but either are not (yet) commercially available [7], [8], or are more aimed at intraoperative navigation [9], [10]. Therefore, a 3-D computerized planning system had to be developed in-house [11], [12]. The aim of this paper is to demonstrate the usefulness of an entirely interactive 3-D planning approach based on a “one object-multiple views” paradigm. A combination of 2-D images, CAD-like implant models and 3-D surface rendered models of bone and prosthesis is highly beneficial to the result of the planning provided the planner is able to interact with them in an intuitive and immediate way. A quantitative and a qualitative validation on a number of patients confirms the added value of this approach. This was especially so when a removable prosthesis was scanned together with the patient to incorporate biomechanical and esthetic considerations into the planning. II. MATERIALS

AND

METHODS

A. Hardware A Somatom Plus CT (Siemens, Erlangen, Germany) is used in spiral mode with a table feed of 1 mm/rotation together with

843

an aperture of 1 mm. Reconstruction is at an interslice distance of 0.5 mm. A maximum height of 40 mm can be scanned with this CT and these settings, but this is sufficient for these structures. The reconstructed images are transferred to the workstation over the intranet of the hospital. The workstation is a Silicon Graphics Octane Maximum Impact with 256-Mb RAM, 4 Mb of texture memory and 4-Gb hard disk.

B. Software The main concept that underlies the whole planning system can be summarized as “one object, multiple views,” meaning that each object will consistently have the same aspect in all windows. This corresponds to what is common for the surgeon, who will not see an intact skin from one angle and a transparent skin from the other. If an object has to change its aspect, e.g., become transparent or invisible, it may do so, but it will change in all windows, so that no extensive bookkeeping is needed to keep track of what is happening in every window. A source of many errors is, thus, avoided. This additional safety is welcome in a system that is used for clinical purposes. The user is working in a virtual environment (the scenery), wherein all objects are located. Toward this scenery, cameras are pointed from random viewing points and with various viewing angles and zoom factors. The windows that are seen on the computer screen are the monitors of these cameras, much as is the case on a film set. Following the philosophy of object-oriented design, the objects themselves know how to “behave themselves:” They monitor their own position and will adapt their properties and aspect accordingly. For example, images from the CT are represented as texture mappings on planes whose locations correspond to the table position during scanning and whose texture corresponds to a slice through the data set that is located as an invisible voxel block underneath the plane. This implies that scrolling through a stack of images is done by altering the spatial position of a plane relative to a block of voxels, whereupon the slice will update its own texture to reflect the appropriate voxels from the CT dataset. Moving the slice out of the voxel block leaves it totally blank, since it then has no underlying voxels. A tilted slice is then analogous to a multiplanar reslice. Additionally, altering the position of a slice will also alter all its intersections with other objects, such as implants and 3-D models. To allow the construction of and interaction with such a virtual environment, the whole planning program is written programming language as a generic library [13] in the C on top of the object-oriented OpenInventor (OI) graphical library, thus, exploiting to the maximum the rendering power of the workstation. This OI library basically uses OpenGL [14], which is a state machine language used to drive the graphic pipeline. The OI file format is used for storage of objects that are derived from OI classes. Every object the surgeon uses during his planning, including the resulting plan, is an equivalent child object derived from an OI parent and can be stored in an OI file. OI objects known as draggers and manipulators act as an interface between the user and the virtual scenery. These

844


(a)

(b)

Fig. 1. (a) Curve with attached reslices and bone model. The implants are in most cases defined on the slice at the arrow. (b) The view of the camera that is attached to the curve.

can constrain interactions to a plane, or only allow rotation around certain points, or modify object properties such as color and transparency. Manipulators such as trackballs alter the orientation of a plane in space, and so provide an interactive multiplanar reslice. Surface models are created separately by the Marching Cubes algorithm [15], [16] as implemented in the visualization toolkit (VTK) [17]. Other functionalities used from VTK are decimation of triangle meshes and connected component selection. Not only surface rendered models are used, volume rendering for an implant neighborhood can be obtained using a sequential set of transparent slices that are aligned to the viewport, so that a correct calculation of the lighting is obtained. C. Planning Steps A stepwise account of the planning procedure with the computerized system is now given. The input consists of the axial images from the CT scanner, and does the reformatting directly on the workstation. To this end a nonuniform rational B-spline (NURBS) curve is added to the scenery, with draggers attached to its control points. The draggers constrain the curve to a plane parallel to the axial plane of the scanner. The user adjusts the curve just as is done on the scanner console for the manual procedure. A set of planes is attached to the curve [Fig. 1(a)]. Each of these planes shows a slice through the data set, thus, immediately providing an interactive multiplanar reslice. If the curve is adapted, the planes will immediately adjust their surface texture. The curve is also used as a guidance path for a camera, whose view is shown in a separate window [Fig. 1(b)]. This camera is aimed

at the slice that intersects the curve along the normal direction and it is moved along it by means of a thumbwheel widget. In contrast to what is done for the manual planning, the planner does the curve reformatting himself and is able to adapt it afterwards depending on additional information, with immediate updates of the attached reslices. Not only does this save valuable time on the scanner console, the planner, thus, also has full control over the angle of the reslices through adjustment of the curve. If the structure in question is a mandible, the user will then move the camera along the jaw and indicate the mandibular nerve, which is sometimes barely visible, by a line. This line interpolates between slices where the nerve or its canal is easily identified (Fig. 2). Once the curve is adequately positioned, and the nerve is delineated, the planner scrolls along it until a suitable location for implant insertion is found. An implant, represented as a cylinder with a central axis, is then defined using two points, commonly named target and entry point. The cylinder has a transparent safety zone around it that indicates a distance of 2 mm, which is the minimal bone thickness that must be kept around it. No other implant may intersect this zone, which is also used to check the distance to the nerve (Fig. 3). Once the first implant is inserted, the angulation is adjusted, and the dimensions adapted to obtain a good (bi)cortical contact. The user then proceeds to insert the next implants. A ruler with a millimeter division is attached to the curve, which aids in finding the nearest possible location for insertion of the next implant. After definition of all implants, the parallelism is checked, which is done by rotating the camera so that it shows frontal and lateral views of all implants (Fig. 6). The alignment of the axes is corrected and the result is checked simultaneously on the 2-D reslices. A 3-D model of the bone surface is then


Fig. 2. Delineation of the nerve on the reslices (inset) creates a line within the 3-D model.

845

of the real world. Several inputs of the human body, such as tactile feedback, are left unused. The disorientation that is often the result can be solved through the use of additional windows and planned paths. The curve that is aligned to the jaw is in fact such a path. The first view, thus, acts as a map whereupon the movement of the camera from the second window is plotted as an adjustable curve. Information overload is another problem inherent to such virtual environments: Normally only the outer surfaces are visible to the surgeon. The inner structures are disclosed stepwise and always with the access path as a guidance. If the internal anatomy is visible the external surface is removed. The two are never present at the same time. Not so during virtual surgery. There is so much information presented simultaneously that a selection of the relevant data for each step is in order. Irrelevant objects can be removed from the visible scenery, which comes down to making them invisible. A problem occurs if these same objects are needed during a later step, since selection of the invisible items is not possible anymore. A special mode is introduced to cope with this problem. The planner has the choice between two desktops, where each desktop contains the objects that are invisible on the other desktop. Selection is then again possible whereupon the object is automatically moved from one desktop to the other. The hardware that is needed for the planning system is expensive and not widely available. Some means, thus, have to be provided to at least view the result of a planning on a standard PC. The views that approximate the intraoperative viewpoints of the surgeon are exported to hyper text markup language (HTML) format, effectively trading interactivity for portability. These views are shown in the operating room using a web-browser (e.g., Netscape) on a standard PC. The 3-D bone surface extracted from the CT corresponds to what is seen upon exposition of the bone, and this is a great aid in positioning and aiming the drill. The browser can also be used as a teleconsultation and training tool and as a demonstrator toward the patient. III. DOUBLE-SCANNING PROCEDURE

Fig. 3. The relation between 2-D reslice, nerve, and implants is clearly seen.

shown together with the implants (Fig. 7). When a patient has a removable prosthesis, the double-scanning technique (see further) is used to incorporate it into the planning, and a model is visualized together with the bone and the implants (Fig. 5). The design process, thus, consists of a loop where 2-D and 3-D are alternatively or simultaneously checked and adjusted. The final result is saved in the OI file format. D. User Interface Aspects A virtual environment has several advantages but also suffers from a user interface that is never as perfect as that

Many patients selected for oral rehabilitation by means of implants already have a removable prosthesis which in many cases reflects optimal tooth position. As we will see, it is highly beneficial to the planning procedure if this prosthesis can be visualized together with the bone structures. One could try to scan the patient while he is wearing his prosthesis. However, most prostheses are made of a resin-like material and, thus, have a very low density, similar to those of the surrounding soft tissue, which makes it impossible to segment out from the CT images by simple thresholding. But even when a prosthesis of sufficient density is present, it is often totally obscured by the streak artifacts caused by fillings in the remaining teeth. In standard jaw acquisitions where bone quality is to be judged these artifacts cause no problems because the scatter lies mainly in the occlusal plane, but for prosthesis design the position of the teeth in the occlusal plane is essential information that should not be lost. Even when only one side of the teeth remains, these are not sufficient to use as an

846


(a)

(b)

(c)

(d)

Fig. 4. (a) Prosthesis scanned separately. The arrows point at the attached markers. (b) The model made from the previous images, showing the same markers as small balls on the structure. (c) The same markers as in the prosthesis scans are visible on the patient.

orientation aid, but they often do create enough artifacts to obscure an eventual radio-opaque template. However, there is a solution to clearly see the prosthesis: A second scan of the prosthesis alone is used to obtain its structure: A clearly visible structure can be extracted for radiolucent objects when they are scanned in air. The contrast with the surrounding medium is now sufficient, but a new problem is created: The position of the prosthesis relative to the patient’s anatomy is lost. To solve this problem, a double-scan technique is used: Small radio-opaque markers are attached to the prosthesis. A first scan of the marked prosthesis shows the prosthesis surface clearly enough together with these markers [Fig. 4(a)], and

allows a good surface rendering of the removable prosthesis [Fig. 4(b)]. Then a second scan of the patient and the intraoral marked prosthesis is made [Fig. 4(c)], showing the location of the markers with respect to the bone, but not the prosthesis itself. A surface model of the bone can, however, be constructed [Fig. 4(d)]. The markers are visible in both sets and so allow the transformation between the two sets to be computed. Once this transformation is known, the scan of the prosthesis can be realigned with the scan of the patient, and both can be inspected together (Fig. 5). Care has to be taken that the computed transformation is orthogonal, however, to avoid deformations of the prosthesis. This is called an orthogonal Procrustes problem, and a solution method is given in [18].


(a)

847

(b)

(c)

Fig. 5. (a) A surface rendered model of the maxillar bone. (b) A model of the prosthesis with fiducials (arrows). (c) The combination of the two registered models.

Fig. 6. The parallelism of the inserted implants is checked from a frontal viewpoint. Note the transparent safety zones around the implants.

Fig. 7. Superimposed bone surface model of maxilla. The configuration of the teeth would be difficult to reconstruct from the 2-D images alone.

IV. VALIDATION

that planner-specific habits would show both in the 2-D and in the 3-D part of the plans, thus, effectively canceling out. The saved plans were compared both by a surgeon and by a prosthodontist. They did not know which plan was the 2-D and which was the 3-D even though the full 3-D functionality was used to inspect the plans in both cases. The following questions were then answered. 1) Which plan is the best for osseointegration? 2) Which plan is the best for biomechanics? 3) Which plan is the best for esthetics? 4) Which plan do you think is the 3-D plan? Moreover, the evaluators motivated certain answers so that a good idea was obtained which features of the system were most useful, thus, resulting in a subjective, but still useful, validation.

The computerized system has significant benefits that make it superior to both the manual and the 2-D-only planning methods. To identify these benefits unambiguously, a quantitative validation was done. A set of 20 patients was planned by a periodontologist, in two stages: The first stage only used the 2-D reformatted images along the curve. No 3-D models or spatial views were used, thus, effectively mimicking the manual planning procedure. However, the scrolling ability along the curve provided more functionality than would be obtained by the manual procedure only, since the intersection of the implants with all reformatted slices could be used. So the “2-D mode” was in fact better than the manual procedure alone and comparable to 2-D computerized systems. The resulting 2-D plan was saved. During the second stage the full 3-D functionality of the system was used to make adjustments to the 2-D plan where deemed necessary. Examples of these corrections are given in the discussion (qualitative validation). The planner did not know at that moment that an intermediate plan was saved and that a study was made, nor that adjustments were expected, thus, avoiding bias. The (eventually adjusted) plan was then saved again as the 3-D plan. It was silently assumed

V. RESULTS

AND

DISCUSSION

An overview of the data is given in Table I. The answers to the questions are given in Table II. A. Quantitative Validation When we look at the results for all the patients together (Table III), corrections are done for ten patients (50%). In

848


PATIENTS

AND

TABLE I IMPLANTS IN

THE

STUDY

No prosthesis # # # #

of of of of

patients planned corrected plans implants corrected implants

11 2 50 2

Prosthesis 9 8 38 19

18% 4%

89% 50%

TABLE II THE ANSWERS GIVEN BY THE EVALUATORS Evaluator 1 Osseointegration? Patients implants

Prosthesis No prosthesis Prosthesis No prosthesis

Biomechanics? Patients implants


Esthetics Patients implants

Prosthesis No prosthesis Prosthesis No prosthesis 3-D?

Patients implants


Evaluator 2

2-D>3-D

2-D=3-D

2-D3-D

2-D=3-D

2-D3-D

2-D=3-D

2-D3-D

2-D=3-D

2-D3-D

2-D=3-D

2-D3-D

2-D=3-D

2-D3-D

2-D=3-D

2-D3-D

2-D=3-D

2-D