Cup position planning for total hip replacement (THR) is a complex task which is influenced by several factors. Whereas aspects like appropriate implant fixation and bone stock preservation are rather evaluated according to intra-operative findings, functional analyses using biomechanical hip models can rely on pre-operative imaging. Due to the wide availability and cost-efficiency of X-ray imaging technology along with the common restriction of biomechanical evaluations to the frontal plane, pre-operative imaging for such purposes is usually limited to AP radiographs. One example is biomechanical optimisation based on the so-called BLB score, which has already been introduced into clinical practice. In this approach, the assumed suitability of potential hip centres of rotation (CORs) is presented to the surgeon by applying colour-coding within the pre-operative AP radiograph. However, to realise the plan, the surgeon has to transfer the 2D positions presented in the radiograph into the 3D surgical site.

We developed a CT-based simulation tool allowing for the generation of 3D bone surface models as well as standardised digitally reconstructed radiographs (DRRs). Within a 3D view, the cup, which is represented as a hemisphere, can freely be shifted in the coronal plane. The 2D point corresponding to the COR defined by the hemisphere is then automatically computed.

In our study, four CT datasets of hips with large bony defects were used. After segmentation 3D bone surface models were generated. These bone surface models were aligned on the basis of the pelvic coordinate system [3], and standardised AP DRRs were computed. BLB score evaluation in intact hips assumes that the central beam passes through the centroid of both hip CORs. As only the contra-lateral hip COR was available due to the defects, a virtual ipsi-lateral COR was obtained by mirroring the contra-lateral hip across the mid-sagittal plane.

Twelve surgeons (divided into two groups of six each according to their experience) had the task to shift the cup such that its 3D position would best match a predefined 2D target position, which was close to the virtual ipsi-lateral COR and displayed as a cross within the standardised DRR. However, the current 2D position corresponding to the current 3D position was not revealed during positioning. Once the user was satisfied with the 3D position, the corresponding 2D position was recorded.

The following results were obtained (mean ± SD across six surgeons of the respective group) for the four patients:

x-error, more experienced: 2.0 ± 6.1; −3.0 ± 5.9; 4.1 ± 4.8; 2.1 ± 5.2; x-error, less experienced: 4.3 ± 4.2; −3.1 ± 1.8; 1.9 ± 4.0; 5.2 ± 4.1; |x-error|, more experienced: 5.2 ± 3.0; 5.4 ± 3.2; 5.5 ± 2.7; 4.3 ± 3.0;|x-error|, less experienced: 4.3 ± 4.2; 3.1 ± 1.8; 3.3 ± 2.7; 5.7 ± 3.3; y-error, more experienced: 12.0 ± 9.1; 0.3 ± 4.3; 6.2 ± 6.6; 1.9 ± 3.2;

y-error, less experienced: 6.1 ± 3.1; 0.8 ± 4.0; 2.4 ± 5.5; 1.4 ± 4.1;|y-error|, more experienced: 12.0 ± 9.1; 3.2 ± 2.6; 6.2 ± 6.6; 3.0 ± 1.9;|y-error|, less experienced: 6.1 ± 3.1; 3.4 ± 1.6; 4.6 ± 3.3; 3.2 ± 2.6;total error, more experienced: 13.5 ± 8.9; 6.6 ± 3.5; 9.8 ± 4.1; 5.4 ± 3.4;total error, less experienced: 8.5 ± 2.7; 4.9 ± 1.5; 6.5 ± 2.5; 6.7 ± 3.8.

Our experimental results show that mental 2D/3D matching for cup positioning in pelvises with bony defects is a difficult task, and that mental 2D/3D matching cannot be expected to yield the correct 3D cup positions corresponding to positions predefined in radiographs. The largest errors were found in the patient with the lowest image quality suggesting that image quality plays an important role. On contrary, experience was not found to be an important factor.

We believe that in clinical practice mental 2D/3D matching between pre-operative radiographs and the surgical site without the help of 3D imaging or special tools would be more difficult than the task given in this study because only small portions of the pelvis would be exposed. Furthermore, as additional aspects of cup positioning would need to be taken into consideration simultaneously, the mental load could be expected to be higher. We conclude that in hips with large bony defects cup positioning based on pre-operative radiographs is highly unreliable without additional computer-assistance or intra-operative imaging. If pre-operative radiographs are needed for functional analyses, combination with 3D image data seems attractive: Firstly, 3D images can easily be used for navigation; secondly, they allow for the generation of highly standardised views, which is essential for comparability across multiple patients.

Future studies relying on more datasets with a wider range of defects could also investigate whether cranio-caudal or medio-lateral positioning errors prevail. This is an interesting question since the BLB score usually is much more specific in the medio-lateral direction than in the cranio-caudal direction, implying that correct 2D/3D matching for the cranio-caudal direction appears less important. In the current study involving only four hips, however, no clear tendency could be observed.

This work has been funded in part by the German Ministry for Education and Research (BMBF) in the framework of the orthoMIT project under grant No. BMBF 01EQ0802/BMBF 01IBE02C.

Consideration of biomechanical aspects during computer assisted orthopaedic surgery (CAOS) is recommendable in order to obtain satisfactory long-term results in total hip arthroplasty (THA). In addition to the absolute value of the hip joint resultant force R the pre- and post-operative orientation of R is an important aspect in the context of the development of a planning module for computer-assisted THA and furthermore for planning of acetabular orientation in periacetabular osteotomy interventions. It is possible to estimate the orientation of hip joint resultant force R for individual patients based on geometrical and anthropometrical parameters. The aim of this study was to examine how far the choice of the mathematical model influences the computational results for the orientation of R in the frontal plane. A further aspect was the comparison of the results with in-vivo data published in the open access OrthoLoad database (www.orthoload.com).

Our comparative study included the 2D-models suggested by Pauwels, Blumentritt and Debrunner as well as the 3D-model suggested by Iglič and three patient datasets from the Orthoload database. As computation of R according to each model relies on standardized X-ray imaging, three anterior-posterior (a.p.) digitally reconstructed radiographs (DRRs) were generated from CT data (x21_x21, x8_x8, x12_x12). The orientation of R was expressed in terms of the angle δ for these three patient individual datasets. The angle δ is defined as the angle between the longitudinal axis and R. The computation results were also compared with in vivo telemetric measurement data from the OrthoLoad database. The following data were used to evaluate R in the frontal plane: the highest load peak of the single leg stance (static conditions) of three patients (EBL, HSR, KWR) respectively in the same manner for planar gait (dynamic conditions) of one patient (KWR). The mean value of the orientation of R under static conditions in single leg stance was calculated in order to get a reference value. For the orientation of R under dynamic conditions δ was calculated by using only the highest peak of three cycles (heel strike to toe off) determined in one single patient (among the three patients involved in the measurements under static conditions) of the database.

The following values of δ were obtained:

Pauwels: 18.26°/20.34°/17.31° (x21_x21/x8_x8/x12_x12) Debrunner: 12.37°/14.30°/12.59° Blumentritt: 5.18°/6.52°/6.14° Iglič: 9.24°/9.01°/9.20°

OrthoLoad database (in-vivo): 28.41°/17.08°/13.32°-static (EBL/HSR/KWR) 16.44°-dynamic (KWR)

The differences in the computational results appear to depend more on the hip model than on the variability of patient-specific geometrical and anthropometrical parameters. The results obtained with in-vivo measurement data are best approximated by using Pauwels' model. The mean values of Pauwels (18.64°), Debrunner (13.09°) and Iglič (9.15°) are a little bit more vertically orientated than the mean value of the static in-vivo results (19.60°). Only Pauwels' model result has a larger angle δ than the in-vivo dynamic result (KWR = 16.44°). By comparing the in-vivo values obtained under dynamic conditions, i.e. gait, (16.44°) with the static in-vivo values of the same patient (13.32°), it could be recognized that the static values are a little bit more vertically orientated than the dynamic result. But both are in the same range as the mathematical models.

The computational biomechanical hip models try to approximate the physiological conditions of the hip joint and the OrthoLoad database represents the physiological reconstructed (artificial) hip joint. Therefore, we think our validation approach is useful for a comparison of the biomechanical computation models.

In contrast, Blumentritt's model outcomes have the largest deviation from the other models as well as from the in-vivo data (static and dynamic conditions). Blumentritt used the weight bearing surface as a reference. He defined it being perpendicular to the longitudinal axis [3]. He postulated that a valid and optimal orientation of R is approximately perpendicular on the weight bearing surface respectively parallel to the longitudinal axis. This approach for validation is questionable because the results show that in the three included and analysed DDR's the orientation is in the mean value 5.95° to the longitudinal axis. It can be concluded that Blumentritt's model assumptions have to be carefully reviewed due to the deviations from in-vivo measurement data.

Among the limitations of our study is the fact that the OrthoLoad database offers only a small number of patient datasets. There is only one dataset for the direct comparison of static (single leg stance) and dynamic (free planar gait) in-vivo measurement data of the same patient included. Furthermore, the individual anatomic geometry data of the patients included in the database are not revealed. Additionally, a source of errors could be an inaccuracy during the data acquisition from the DRR.

Further research seems to be recommendable in the context of implementing a biomechanical hip model in a planning module for computer-assisted THA or periacetabular osteotomy interventions, respectively. Sensitivity analyses and parameter studies for different mathematical models using a multi-body-simulation system are objectives of our ongoing work.

Biomechanical considerations are relevant to cup positioning in total hip replacement (THR) to optimise the patient-specific post-operative outcome. One goal is to place the hip centre of rotation (COR) such that parameters characterising the biomechanics of the hip joint lie within physiological ranges. Different biomechanical models have been developed and are based on exact knowledge about muscle insertion points whose positions can be estimated on the basis of bony landmarks. Therefore, accurate landmark localisation is necessary to obtain reliable and comparable parameter values.

As most biomechanical considerations are limited to the frontal plane, landmark localisation relying on standardised pre-operative radiographs has been established in clinical practice. One potential drawback of this approach is that user-interactive landmark localisation in radiographs might be more error-prone and subjective than localisation in 3D images. Therefore, we investigated the possibility of increasing the reproducibility of interactive landmark localisation by providing 3D localisation techniques. As the so-called BLB score based on Blumentritt's biomechanical hip model has already been introduced into clinical practice as a criterion for cup position planning, we examined the anatomical landmarks involved in BLB score evaluation. We developed a CT-based simulation tool allowing for the generation of 3D bone surface models and standardised digitally reconstructed radiographs (DRRs). Correspondences between points in the 2D DRR and rays in the 3D bone surface model are automatically established and optionally visualised by the tool.

Two modes of landmark localisation were examined: In the 2D-mode, only AP DRRs were displayed, and the users had to localise the landmarks by clicking within the DRR image. In the 3D-mode, additionally the arbitrarily rotatable bone surface models together with the aforementioned 2D/3D correspondences were visualised. The user could then choose between landmark localisation by clicking either within the DRR image or within the 3D view. In either case, the 2D landmark positions within the DRR were recorded.

The participants were given both an example AP pelvis radiograph with highlighted anatomical landmarks and the following landmark descriptions from the user's manual (v2.06) of the mediCAD software (Hectec GmbH, Landshut, Germany): P4: ca. 3cm distal lesser trochanter minor (in the imagined direction of pull of the rectus femoris muscle towards the medial upper edge of the patella); P5:lateral, most proximal edge of the trochanter major; P6: most cranial edge of the sclerotic area; P7:spina iliaca anterior inferior; P8/P9:most lateral/cranial point of the wing of the ilium.

(P1 and P2 are only needed to define the position of the mid-sagittal plane, and P3 is the pre-operative COR. Due to correct radiograph standardisation, we assumed this plane and P3 to be known prior to landmark localisation.)

Thirteen surgeons repeated the experiments on four hips (CT datasets of two male patients).

The following results were obtained (SD of relevant coordinates obtained with 2D localisation vs. SD of those obtained with 3D localisation) in the first patient (left hip: 1L; right hip: 1R) and the second patient (left hip: 2L; right hip: 2R):P4: 6.3 vs. 9.0 (1L); 6.7 vs. 5.6 (1R); 9.0 vs. 11.1 (2L); 7.1 vs. 8.6 (2R); P5: 4.4 vs. 2.8 (1L); 3.1 vs. 3.1 (1R); 4.3 vs. 2.4 (2L); 4.7 vs. 4.1 (2R); P6: 4.8 vs. 3.8 (1L); 2.9 vs. 2.8 (1R); 3.7 vs. 5.2 (2L); 6.9 vs. 3.5 (2R); P7: 12.2 vs. 6.1 (1L); 12.1 vs. 3.7 (1R); 7.6 vs. 4.6 (2L); 6.2 vs. 4.5 (2R); P8: 1.2 vs. 2.8 (1L); 2.0 vs. 2.6 (1R); 1.5 vs. 2.1 (2L); 2.0 vs. 1.6 (2R);P8: 4.1 vs. 2.1 (1L); 7.3 vs. 3.9 (1R); 1.6 vs. 2.6 (2L); 4.1 vs. 3.2 (2R).

The greatest differences in reproducibility were observed in P7, which was barely distinguishable in the radiographs and, hence, showed very low reproducibility only for the 2D-mode. P4 showed low reproducibility in both modes due to its vague description and the relatively small portions of the femurs contained in the CT-scanned volume. In P9 the low reproducibility obtained with the 2D-mode might be partly explained by truncation artefacts present in the DRRs.

Although our study needs to be extended to more datasets, we conclude that the availability of 3D data allows for higher landmark localisation reproducibility when compared with the conventional X-ray-based approach, which has additional drawbacks: Standardisation of X-ray imaging, which is necessary to retain comparability of biomechanical parameter values determined in different patients, is hard to achieve; specifications e.g. concerning the central beam may be met only after acquiring several radiographs. Moreover, once a 2D target cup position is defined based on the 2D biomechanical analyses, the transfer of this position into the 3D surgical site is difficult without additional 3D imaging.

Hence, the use of 3D imaging and 3D landmark localisation techniques seems more promising for cup positioning based on biomechanical models, which, however, need validation.