Computer-aided planning of reconstructive surgery of the innominate bone: Automated correction proposals

Abstract

Objective: In cases of complex reconstructive surgery of the innominate bone, it is difficult to draw up a good surgical plan; manual planning of a 3D reconstruction is highly demanding and time-consuming. This paper presents and validates methodology to automatically generate 3D correction proposals for reconstructive surgery of the innominate bone, and illustrates its use with clinical applications.

Materials and Methods: The developed Matlab® methodology starts from CT-based outer surface representations of the patient's bone, which allow straightforward mirror and matching implementations for automated reconstruction procedures. The validation on 9 cadavers was two-fold: the geometrical deviations of the intact original with respect to the reconstructed surface meshes were assessed, and the characteristics of both original and reconstructed acetabular cup regions were determined.

Results: Eighteen healthy and thus, it was assumed, spherical acetabula were automatically reconstructed with mean accuracies of 3.2 ± 2.2 mm, 0.1 ± 1.0 mm and 3.8 ± 2.9° for the hip joint centerpoint, joint radius and cup orientation, respectively. As a demonstration, a triflange cup acetabular implant was rapidly designed, starting from the correction proposal.

Conclusions: A highly automated, computer-aided approach to surgical planning for pelvic bone defects was developed and sample applications demonstrated. Validation results for healthy acetabula were superior to those obtained in real surgery. The generated virtual correction proposals can be used as targets in surgical planning and cup navigation applications, or in the design of customized implants with complex shapes.

• Computer assisted orthopaedic surgery
• surgical planning
• pelvis reconstruction
• triflange cup acetabular implant

Introduction

Defects and malformations of the innominate bone are diagnosed on a daily basis and reconstructive surgery is often inevitable. Common examples are acetabular revisions with excessive bone loss, possibly resulting in pelvic discontinuity [1], [2], and osteotomies of the innominate bone [3–6]. Currently, multiple imaging modalities such as X-rays and CT-based visualization software, along with image analysis tools such as 2D (CE angle) and 3D (coverage) parameters and their respective reference values, are at the surgeon's disposal when planning bone reconstruction or repositioni [7], [8]. Additionally, various bone defect classifications (as in the system developed by the American Academy of Orthopaedic Surgeons [AAOS]) may indicate a possible surgical approach or a preferable implant type [1]. Nevertheless, these analysis and planning facilities do not directly provide an actual correction proposal which can be used to set up accurate navigatio [9], [10] or robot [11] systems for the placement of standardized implants; neither do they directly provide a customized implant shape. Further manual input is required to plan a reconstructive surgery of the innominate bone. Unfortunately, such planning is user-dependent, highly demanding and time-consumi [12], [13].

This paper presents methodology to automatically generate 3D correction proposals for bone reconstructive surgeries, such as reconstruction of the innominate bone. The developed Matlab® (The MathWorks Inc., Natick, MA) methodology starts from CT-based outer surface representations of the patient's bone, obtained from a generally developed filter and mesh base platform [14–16]. These representations allow straightforward implementation of automated reconstruction procedures, which can serve either as a specialized diagnosis tool, or as a valuable correction proposal. This is illustrated by three clinically very distinct applications: correction proposals for a total hip prosthesis (THP), hip dysplasia, and tumor surgery. A quantitative validation of the presented procedure is performed for the acetabulum, which is the most critical and complex-shaped region of the innominate bone.

The automatically generated virtual bone correction proposals can be used as targets in cup navigation applications or in the design of complex-shaped implants. The latter application is illustrated for a triflange cup acetabular implant desi [12], [13]. This application demonstrates that, using the developed methodology, 3D planning of complex bone reconstructions becomes user-independent, relatively undemanding and swift.

Materials and methods

Preparation: Filter and mesh base platform

The bone contours are extracted from the patient's CT data by means of a quick grey value segmentation in Mimics® (Materialise N.V., Leuven, Belgium) (for example), followed by application of a Marching Squares algorithm [17] (Figure 1, left). Then, an extended filter procedure–as previously developed by the authors [14]–only retains contour information representing the outer surface, as more specific internal contours, fragmented contours, internal loops and shape irregularities are removed, tailoring the image for the application (Figure 1, center).

Figure 1. General filter and mesh base platform [6], illustrated for an innominate bone. Bone contours, which are extracted from a CT scan (left), are filtered for outer contour information (center) and converted into an outer surface triangulation mesh (right). [Color version available online.]

A surface mesh is built from the spline set, created from the contours by the above-mentioned filter procedure. Subsequently, a grid of quadrilaterals is created after selecting a fixed number of points on each spline. Each branch is processed separately and an intermediate layer of triangles ensures a smooth transition of the branches (Figure 1, right). Finally, the surface mesh is closed at the ending contours, then converted to STL (Standard Triangulation Language) file format. A typical innominate bone mesh, built from contours with 2 mm of interslice distance and sampled with 50 points per spline, contains approximately 6000 mesh points (12 000 triangles).

Automated reconstruction procedure

Using the filter and mesh base platform, filtered surface meshes are built for both the affected and intact contralateral innominate bones. The user then specifies the defect range by indicating two points on the mesh (Figure 2a). The data within this range is unreliable and will therefore be deleted, while the remaining intact bone regions of the mesh are automatically mirrored and rigidly matched [18] to the intact contralateral mesh. Subsequently, contralateral contour information is retrieved by reslicing the intact surface mesh according to the slice direction of the matched affected mesh, and back-transformed (Figure 2b). Finally, a slight smoothing is applied to the reconstructed mesh in order to remove noise and, if applicable, to enhance the shape of the transition between the reconstruction and the healthy mesh region (Figure 2c). This is done in the Magics® software (Materialise N.V., Belgium) (for example), by applying a global smoothing to the mesh with a user-defined smoothing percentage (typically 50%).

Figure 2. Automated bone reconstruction: acetabulum. The user marks the region to be reconstructed (a), e.g., a fractured innominate bone region with metallic CT artefacts. The procedure then performs an arbitrary mirroring and matching operation on the remaining bone parts. In this position, the intact contralateral innominate bone is resliced to retrieve intact geometrical information. The obtained contour information is then back-transformed and incorporated into the original contour set (b). Finally, the surface mesh is updated (c). The latter directly serves as a surgical goal, e.g., for positioning standardized cups or shaping bone grafts, or as a starting point in the design of customized implants. [Color version available online.]

Clinical examples

Three simulation cases were performed to test the developed automated reconstruction methodology, and to illustrate the benefits in distinct surgical situations. These simulations concerned the planning of a total hip prosthesis, diagnosis of hip dysplasia and reconstructive planning, and the planning of bone tumor surgery. The first and third cases were retrospective, allowing a direct comparison with current planning methodology.

Validation for acetabula

Nine cadaver specimens (Caucasian; age 85 ± 10 years; 1 male, 8 females) with healthy, and thus approximately symmetrical, innominate bones were scanned with a Somatom Sensation spiral CT scanner (Siemens, Erlangen, Germany) (cubic voxel size: 1 mm³). Subsequently, the above-described automated reconstruction procedure was applied to the 18 acetabular regions. As the cadavers were assumed to exhibit contralateral symmetry, it was expected that the reconstruction proposal would be equivalent to the original situation.

The validation was two-fold. On the one hand, the geometrical deviations of the original with respect to the (unsmoothed) reconstructed surface meshes were assessed using the Focus Inspection® software (Metris N.V, Belgium) (Figure 5a). This software quantifies and visualizes the distance offsets between two surfaces meshes. On the other hand, the characteristics of both reconstructed (rec) and original (orig) acetabular cup regions were determined by interactively assessing the acetabular rims, retrieving the enclosed acetabular mesh regions, and fitting least-squares spheres into these acetabula (Figure 3). From these objects, the hip joint centers, diameters and acetabulum orientations (least-squares plane normal) were derived. For both left and right sides X of each specimen i, the discrepancies Δ are determined as the Euclidean distance d between the hip joint centers c_iX, the difference in sphere radius r_iX, and the enclosed angle between the normal vectors. The latter was calculated as the arccosine of the scalar product of the normalized normal vectors

Figure 3. Acetabular cup characteristics. A least-squares sphere is fitted in the acetabulum. From this object, the hip joint center and diameter are assessed. The acetabulum orientation is defined as the plane normal from a least-squares plane through the acetabular rim points. The latter are visualized as black dots. [Color version available online.]

It must be mentioned that, in reality, the symmetry assumed for the innominate bones is only approximate. Therefore, the cadaver specimens were selected with the utmost care; the values for standard deviation and range for the validated acetabular cup characteristics will confirm the validity of the cadaver selection.

Results

Clinical examples

The first case concerns a reconstruction proposal for revision THP cup placement in combination with correction of an acetabular fracture. The acetabular region, as shown by a volume STL rendering in Figure 2a, is unreliable due to loose bone fragments or metallic CT artefacts. The resulting reconstruction proposal seen in Figure 2c directly serves as a surgical goal.

In Figure 4a, the procedure is applied to a dysplastic hip. By comparing the original and reconstructed meshes, the geometrical distortion could be assessed immediately. The proximal acetabular region is extremely flattened; the proposal suggested a femur displacement of 2 cm in the distal direction.

Figure 4. Automated correction proposals for the innominate bone: (a) planning and/or diagnosis of hip dysplasia; (b) tumor reconstructive surgery. For each case (from left to right), the original, original and correction, and correction proposal are presented. An arrow indicates the location of the defect/deformity in the original bone mesh; in (a), a pair of horizontal lines indicates the required vertical displacement of the patient's femur. [Color version available online.]

Finally, Figure 4b presents a right innominate bone with a tumor in the body of the ilium. Although a surgeon can preoperatively identify the tumor tissue by analyzing the CT grey values, no reconstruction plan results from this analysis. In contrast, the developed methodology directly generates a proposal by incorporating information from the healthy contralateral pelvis, and presents it to the surgeon for visual inspection/interpretation.

All three reconstruction procedures took approximately 7 minutes. The preceding filtering of both innominate bone meshes lasted 30 minutes (these calculation times reflect use of a 2-GHz Celeron® processor, with 512 MB of RAM, and uncompiled Matlab® code).

Validation for acetabula

Entire mesh comparison. Figure 5a presents a Focus Inspection® deviation color map of one of the 18 cadaver innominate bones. For this innominate bone, the distance deviations between the original and reconstructed mesh are small and mainly located at the transition regions, not in the acetabular region itself. The histogram of the distance deviations presented in Figure 5b generalizes these findings. It contains two distinct regions: a central peak mainly designates the unchanged mesh parts such as the ilium and ischium, while the extremities indicate maximal deviations of 4 mm in only a few points (out of 6000 mesh points).

Figure 5. Innominate bone mesh distance deviations after automated acetabulum reconstruction: (a) color plot; (b) deviation histogram. A typical innominate bone mesh, built from contours with 2 mm of interslice distance and sampled with 50 points per spline, contains approximately 6000 mesh points. [Color version available online.]

Acetabular cup characteristics

Table I presents the statistics for the acetabulum characteristics illustrated in Figure 3. On average, the sphere centerpoints collide within 3.2 ± 2.2 mm, the sphere radii are similar within 0.1 ± 1.0 mm, and the direction normal to the acetabular rim deviates by 3.8 ± 2.9°.

Table I.  Acetabular cup deviation statistics for left (right) corresponding original and reconstructed acetabula. (n = 18, i = specimen, X = specimen side (left or right); Δc_iX, Δr_iX and Δn_iX, are the discrepancies in centerpoint, radius, and normal, respectively. See Figure 3 and formulae 1–3 for a detailed description).

	Mean ± σ	Max
ΔciX (mm)	3.2 ± 2.2	7.1
ΔriX (mm)	0.1 ± 1.0	3.3*
ΔniX (°)	3.8 ± 2.9	8.3

σ = standard deviation. * = as absolute value.

Discussion

Demand for complex and reconstructive surgery of the innominate bone will increase during the coming decades. The need for efficient surgical planning features will therefore become proportionally important.

Firstly, increasing numbers of patients have been treated at a young age for different types of osteoarthritis by total hip replacement. However, aseptic loosening continues to represent the most common mechanism of failure in cemented and non-cemented total hip arthroplasty. Previous studies have shown a 12.5% incidence of acetabular component loosening at 5.75 years [19]. Ritter et al. [20] noted a 39% incidence of complete radiolucency around the acetabular component; a 4% incidence of migration and loosening of the acetabular component is very often associated with aggressive osteolysis resulting in significant bone loss from the innominate bone [21].

Secondly, the innominate bone can also be destroyed by primary or metastatic malignant disease. When there is bone destruction and medial or superior migration of the femoral head, total hip arthroplasty is required. In these cases, one must realize that there may not be sufficient bone for conventional sealing of the acetabular component. Reconstruction with poly-methylmethacrylate (PMMA) often seems the obvious solution, although previous reports have documented the inadequacy, in terms of duration of fixation, of PMMA alone for reconstructing lost bone or sealing the acetabular cup in a migrated positio [22], [23].

Thirdly, total hip arthroplasty is also indicated as an acute primary treatment for complex acetabular fractures and as a sealing procedure after completing initial treatment by traction or internal fixation. This operation is a very severe intervention associated with a high complication rate. For new fractures, the goal of the intervention is the restoration of the acetabular dome and the medial and lateral column. This is rarely possible when a conventional acetabular cup is used [24–26]. Osteopenia could also make an operative reconstruction very difficult [26–28].

Fourthly, Romness and Lewallen [29], using Stauffer's criteria [30], observed an incidence of radiographic loosening of 52.9%, an incidence of symptomatic loosening of 27.5%, and an acetabular revision rate of 13.5%. Using the Kaplan–Meier method, they projected the 10-year revision rate to be 18.2%.

Complex reconstructive surgery of the innominate bone requires a good 3D surgical plan. Unfortunately, this planning is, with currently available manual computer-assisted planning tools, both demanding and time-consumi[1], [12], [13]. The presented methodology demonstrates that a correction proposal for pelvic reconstructive surgery can be attained automatically and swiftly. Starting from both the affected and intact bone surface meshes, an automated mirroring step is performed. No user interaction is required to outline the optimal plane of symmetry.

The reconstruction procedure requires the contralateral innominate bone to be intact. Since in practice this will not always be the case, a database of reference bones is currently set up and linked to the presented methodology.

A quantitative validation of the presented reconstruction procedure has been performed for the acetabulum, which is the most critical and most complex-shaped region of the innominate bone. A retrospective analysis of THP cases has not been performed, but could be favorable in further endorsing the current validation results. Nevertheless, it should be mentioned that such retrospective analysis could be ambiguous. On the one hand, similar results could prove the validity or redundancy of the proposal tool (not taking into account the time spent in manual surgery planning). On the other hand, differing proposals should be judged on the basis of their long-term clinical outcomes, which are unavailable for virtual proposals. The currently adopted validation methodology bypasses such possible ambiguity.

Results for the quantification parameters for the acetabular deviations, i.e., hip joint size, center location and acetabulum orientation, are satisfactory for the intended applications. They are superior to those obtained during real surgery [9], [10]. The values for standard deviation and range for the acetabular cup characteristics indicate that the cadaver specimens were indeed symmetrical, allowing for small natural deviations.

The clinical applicability of the developed procedure as a diagnostic tool and creator of surgical targets has been proved by three very distinct simulation cases, namely total hip prosthesis planning, reconstructive surgery planning, and hip dysplasia detection and reconstruction planning.

Application: Triflange cup acetabular implant design

The automatically generated virtual bone correction proposals can be used as targets in cup navigation applications, or in the design of implants of complex shape. The latter application is illustrated for a triflange cup acetabular implant desi[12], [13], [31]. This custom-made implant is indicated for preoperative innominate bone deficiencies classified as AAOS Type III (combined deficiency) or IV (pelvic discontinuity), or Paprosky type IIIB. A hemispherical cup is rigidly connected with large customized flanges to the ilium, ischium and pubis. Currently, the design of such a triflange cup implant is difficult and demanding, even with the use of–mainly manual–CAD software. Using the above-described methodology, however, 3D planning of complex bone reconstructions can become user-independent, relatively undemanding, and swift.

Complex-shaped implants can be extracted directly from the automatically corrected bone surface models. A least-squares sphere is fitted in the acetabulum from the correction proposal, by the same method as illustrated in Figure 3. The acetabular region of the reconstruction proposal is then swapped with a spherical cup region, with a standard diameter that is smaller than that of the fitted sphere (Figure 6a). Subsequently, the user outlines the triflange cup implant on the resulting mesh and retrieves the enclosed region by using a polyplane cutting tool (Figure 6a). If a volume implant is preferred, a bone piece is cut out from the surface mesh and connected to the triflange surface (Figure 6b).

Figure 6. Triflange cup design extraction. A direct link between the reconstruction proposal and the implant design is provided by additional surface mesh cutting tools. This results in either a triflange cup plate implant (a) or a triflange cup volume implant (b). [Color version available online.]

Further research will focus on establishing an efficient link between the aforementioned implant design and production. The resulting implant design, either a plate structure or a monobloc, will be the input for specific plate formation [32], milli [13], [33] or rapid prototyping [34] procedures.

Conclusion

The methodology for automatically generating 3D correction proposals for reconstructive surgery of the innominate bone has been described, validated and illustrated. Validation results for 18 healthy acetabula indicate that the developed simulation methodology restores normal anatomy with an accuracy which is superior to that obtained in real surgery. On the one hand, three clinically very distinct applications–total hip prosthesis, hip dysplasia and tumor surgery - illustrate the strength of the procedure as a diagnostic tool and as a creator of surgical targets. The reconstructed models can, for example, be used to set up surgical navigation systems or to shape graft materials. On the other hand, the example of the triflange cup acetabular implant design demonstrates the technical and medical benefits conferred by the developed methodology: rapid, accurate, and user-independent computer-aided pre-operative implant design.

Acknowledgments

We wish to thank Prof. Dr. P. Vanderschot (Department of Surgery, University Hospital Gasthuisberg, K.U. Leuven, Belgium) and Dr. I. Samson (Department of Surgery, University Hospital Pellenberg, K.U. Leuven, Belgium) for providing the CT data of the hip dysplasia and innominate bone tumor, respectively, and the Materialise company for putting at our disposal the Mimics® and Magics® software. This research is funded by a PhD grant from the Institute for the Promotion of Innovation through Science and Technology in Flanders (IWT-Vlaanderen).