Statistical shape modeling of the large acetabular defect in hip revision surgery

The assessment of three‐dimensional bony defects is important to inform the surgical planning of hip reconstruction. Mirroring of the contralateral side has been previously used to measure the hip center of rotation (CoR). However, the contralateral side may not be useful when diseased or replaced. Statistical Shape Models (SSMs) can aid reconstruction of patient anatomy. Previous studies have been limited to computational models only or small patient cohorts. We used SSM as a tool to help derive landmarks that are often absent in hip joints of patients with large acetabular defects. Our aim was to compare the reconstructed pelvis with patients who have previously undergone hip revision. This retrospective cohort study involved 38 patients with Paprosky type IIIB defects. An SSM was built on 50 healthy pelvises and used to virtually reconstruct the native pelvic morphology for all cases. The outcome measures were the difference in CoR for (1) SSM versus diseased hip, (2) SSM versus plan, and (3) SSM versus contralateral healthy hip. The median differences in CoR were 31.17 mm (interquartile range [IQR]: 43.80–19.87 mm), 8.53 mm (IQR: 12.76–5.74 mm), and 7.84 mm (IQR: 10.13–5.13 mm), respectively. No statistical difference (p > 0.05) was found between the SSM versus plan and the SSM versus contralateral CoRs. Our findings show that the SSM model can be used to reconstruct the absent bony landmarks of patients with significant lysis regardless of the defect severity, hence aiding the surgical planning of hip reconstruction and implant design.

reconstruction of patient anatomy. Previous studies have been limited to computational models only or small patient cohorts. We used SSM as a tool to help derive landmarks that are often absent in hip joints of patients with large acetabular defects.
Our aim was to compare the reconstructed pelvis with patients who have previously undergone hip revision. This retrospective cohort study involved 38 patients with Total hip arthroplasty (THA) is a widespread orthopedic procedure used to restore hip joint function when severe acetabular defects are present. 1 The management of severe acetabular bone loss is a challenging task in revision THA. Multiple treatment options have been proposed such as porous tantalum acetabular components with or without structural allograft or metal augments, 2,3 standard cage reconstruction with iliac or ischial screw fixation, 3,4 and cupcage constructs. 5 However, these techniques often come with unsatisfactory results and present a high revision rate due to implant failure. 6 Custom-made acetabular implants can overcome these challenges by fitting the implant to the residual host bone, playing a crucial role in complex reconstruction surgery. 7 The use of preoperative planning and patient specific instrumentation allows for an accurate positioning of the acetabular components for a better performance, reducing the need for revision surgery. 8,9 Mirroring of the healthy contralateral hemipelvis has been previously used to measure the hip center of rotation (CoR). 9 However, this technique is affected by human anatomy asymmetry and cannot be applied when the contralateral hemipelvis is pathological or metal work is present. 10 Literature shows how pelvic Statistical Shape Models (SSMs) have mainly been used for automatic bone segmentation purposes [11][12][13] and also to reconstruct pelvic bone defects, [14][15][16] however, these studies were limited to computational models only (no patients involved) or patient cohorts as small as two or eight patients. 15,16 We aimed to better understand whether SSM could be used to aid the reconstruction of important bony landmarks that are absent in hip joints of patients with large acetabular defects. Our objective was to compare the CoR of the SSM with patients who had previously undergone hip revision surgery for large acetabular defects planned without the SSM technique.

| Study design and outcome measures
This was a retrospective cohort study involving 38 patients with Paprosky type IIIB defects. Type III acetabular defects present a major destruction of the acetabular rim and the teardrop. They also show severe lysis of the ischium, resulting in superomedial component migration. 1 An SSM was trained on the basis of 50 healthy pelvises and was used to virtually reconstruct the native pelvic morphology for all 38 cases. Within the cohort, 18 of the 38 patients had healthy contralateral sides. The SSM was then compared with the preoperative computerized tomography (CT)-based plan for all patients whose surgery was planned without the SSM technique.
Study design is shown in Figure 1.

| Test set
The test set consisted of 38 patients with Paprosky IIIB defects, 18 females and 10 males. Their mean age was 69 years (standard deviation [SD] = 10.64). Their preoperative CT scans were segmented using the same software by mean of a combination of manual and automatic segmentation tools. The 3D virtual plans of the 38 cases were used to test the statistical model, and the contralateral hemipelvises of patients that did not present an implant were investigated. Out of the 38 patients, 2 did not undergo revision surgery and only the virtual plan was available.

| Validation set
Ten healthy hemipelvises formed the validation set. Five females and five males were selected from the training set and used to validate the virtual reconstruction process.

| SSM
The 50 pelvises that formed the training set were initially aligned to ensure a fixed pose of the data set. A mean shape of the hemipelvis was then registered to each individual hemipelvis using a point mapping technique. The mean shape served as the reference object and provided the locations for the points to be mapped. Principal component analysis was applied on the data set to investigate the correlations between the mean shape and the model set using three F I G U R E 1 Flowchart of the study design. CoR, center of rotation.
De ANGELIS ET AL. | 2017 modes of variation. Assuming that the healthy parts of the diseased hips that formed the test set could predict the shape of the anatomy, the acetabulum was removed and the remaining parts were used to build the SSM. The healthy parts of the diseased hip were aligned to the best fit model which was generated from the mean shape and the dense data points of the individual geometries.

| Image analysis
The CoR of the SSM was calculated using a sphere matching technique. 17,18 The same method was used to compute the CoR of the diseased hip, the plan and the healthy contralateral side.
First, the difference in CoR between each diseased hip and its respective SSM was calculated to quantify the severity of the defect. Second, the difference in CoR between the SSM and the plan was computed. Lastly, the difference in CoR between the contralateral healthy side and the SSM was calculated. To provide XYZ interpatient relevance, the coordinate system was changed with respect to the anterior pelvic plane (APP) using the anatomical definitions. 19 The APP was defined using three points: (1) the right anterior superior iliac spine, (2) the left anterior superior iliac spine, and (3) the pubic tubercle. Using these reference points, the plane was created and the centroid distances were calculated. X corresponded to the sagittal plane, Y to the axial plane, and Z to the coronal one.

| Statistical analysis
Statistical analysis was performed using GraphPad Prism (version 9.1; GraphPad Software) to investigate statistical significance between the groups. Normal distribution of the data was checked by means of the D'Agostino normality test for each group. The data were determined to be nonnormally distributed. A Mann-Whitney U test for nonparametric independent values was carried out. The significance level was set at p < 0.05.

| Validation
The SSM built using the training set was used to virtually reconstruct the hemipelvises of 10 healthy patients. Their CoR was measured using the sphere matching technique described in Section 2.3. 17,18 The difference in CoR with respect to their corresponding SSM was calculated to test the performance of the model.

| RESULTS
The image analysis and validation results are reported. We present the discrepancy in CoR between the diseased hip and the SSM, the plan and the SSM, and the healthy contralateral side and the SSM.

| Statistical analysis
A significant difference (p < 0.0001) was reported between defect versus SSM and plan versus SSM difference in CoR. A statistical difference (p < 0.0001) was also found between defect

| Validation
The median difference in CoR between the healthy hemipelvises and their SSMs was 5.69 mm (IQR: 6.37-3.77 mm). Figure 10 shows the validation results. An average case is illustrated in Figure 11.
The largest displacement in CoR was found in the coronal plane.

| DISCUSSION
The preoperative assessment of large acetabular defects such as Paprosky type III is challenging. 20    The preoperative and postoperative radiographs further investigate the use of SSM when the contralateral anatomy is diseased or replaced. The patient presented with a right failed hip implant. As an implant was also present on the other side, no contralateral anatomical landmark could be used as a reference to calculate the CoR of the side of interest. Using SSM over standard 3D preoperative planning would have allowed us to derive the anatomical landmarks absent in the failed hip joint and reconstruct the CoR of this complex case ( Figure 8).
We acknowledge limitations. First, as the aim of the surgical plan is not always to restore the native CoR, seven patients had to be excluded from the plan versus SSM comparison, and corresponding statistical analysis. Lastly, a larger patient cohort should be used to further assess whether the model is able to cope with different types of severe acetabular defects or deformities. Other factors, such as age, could also be investigated and pelvises of the younger population could be incorporated into the training sets.

| CONCLUSION
Planning hip reconstruction in patients with large acetabular defects is difficult due to the lack of anatomical landmarks, as a result of the deformed anatomy and contralateral hip being often replaced. In addition, the presence of the failed implants creates metal artifacts that obscure bony readings from CT scans making the 3D reconstruction challenging. SSM can be used to overcome these limitations. We present the first study that applies an SSM tool to a cohort of 38 patients with Paprosky IIIB defects.