MRI-based static and functional assessment of complex hip deformities in comparison with CT: a validation study

Our understanding of hip pathobiomechanics has evolved over the past decades and has led to advances in joint-preserving surgery [1]. Accurate analysis of hip morphology by advanced imaging modalities had an important role in this development [1]. Besides determination of conventional radiographic parameters, radiologists can use computer-assisted animation to display the virtual range of motion (ROM) and detect possible sources of intra- and extra-articular impingement [2‐4]. This information helps the surgeon to better understand the deformity, especially in complex cases with combined deformities, and plan the best surgical strategy [1]. Over the last decade, the technique for motion analysis has improved from using a fixed predetermined center of rotation to the computation of an acetabular and femoral sphere, which maintains a dynamic center of rotation with an equidistant joint space and thus makes virtual ROM more precise [3]. The algorithms for virtual ROM analysis have been validated on cadaveric models by comparing actual and virtual ROM based on computed tomography (CT) data [2, 5].

Previously, only CT was used for virtual ROM measurement due to its superior contrast between bone and soft tissue and spatial resolution. Despite the radiation hazard, CT continues to be regarded as the gold standard for quantifying bone abnormalities [6]. Therefore, reliable alternative imaging techniques not using ionizing radiation are desirable, especially as many of the patients requiring hip-preserving surgery are children or young adults. State-of-the-art high-resolution 3D magnetic resonance imaging (MRI) sequences fulfill these requirements and can be used to acquire imaging data for virtual motion analysis. MRI has been shown to be comparable to CT for the static measurement of the femur [7] and acetabulum [8, 9]. With the advent of improved semiautomatic segmentation algorithms for volumetric MRI datasets, it has become possible to use virtual ROM analysis in clinical practice [4]. Therefore, the purpose of this study was to assess the feasibility of MRI-derived quantification of hip joint morphology and virtual ROM analysis in patients with hip deformities using CT as a standard of reference.

Acetabular parameters showed excellent ICCs for intermethod correlation with “almost perfect” agreement for anteversion, ASA, sourcil angle, and acetabular coverage (Table 2). The Bland–Altman plots revealed an even distribution of differences around the zero line with 95% confidence intervals (95% CIs) of 0.08° ± 3.46°, − 0.64° ± 3.27°, and up to 0.40° ± 2.36° for acetabular anteversion, sourcil angle, and acetabular sector angles, respectively (Fig. 1a–c). Linear regression analysis detected no significant proportional bias, and the TOST showed no significant occurrence of mean differences ≥ 2° for any of the comparisons (Table 2).

Femoral parameters showed good correlation as well. However, with a difference ranging from − 2.5° to 1.3° the comparability of the alpha angle was not as good as that of the other parameters. Still, all ICCs for intermethod correlation showed almost perfect matching except for the alpha angle at 9 o’clock, which demonstrated only ‘substantial agreement’ (Table 2). The Bland–Altman plot also revealed a wider 95% CI and an even distribution around the zero line (Fig. 2). The TOST showed no significant occurrences of  ≥ 2° differences for the alpha angle, except at 12 o’clock (Table 2).

For femoral torsion and the CCD angle, the mean absolute difference was 0.35° ± 3.12° and − 0.18° ± 1.64°, respectively. The Bland–Altmann plots are shown in Fig. 1d, e. An ‘almost perfect’ intermethod agreement was found for femoral torsion (ICC 0.97) and CCD angle (ICC 0.98) (Table 2). Detailed analysis of the cases with differences greater than 3° in femoral torsion revealed an incorrect position of orientation points in either CT or MRI in 4 of 37 hips. To illustrate, two cases are presented in Fig. 3. Linear regression analysis detected no significant proportional bias (Table 2). No significant influence of a time interval  > 5 min between VIBE and HASTE sequences on discordant results between MRI and CT was found for femoral torsion (p = 0.91, Table 3) (Fig. 4).

A total of 370 ROM analyses were performed in 37 hips (10 different movements per hip). In 78.4% (290/370) of the analysed movements, neither CT nor MRI detected impingement in the predefined physiological range of motion (Table 1). In 18.6% (69/370) of the movement analyses, both CT and MRI detected impingement in the physiological range of motion with a mean difference of − 0.1° ± 2.4° (− 21° to 13°). In 2.2% (8/370) of cases, only CT motion analysis, but not MRI analysis, detected impingement in the physiological range of motion. In 0.8% (3/370) of cases, impingement was detected by MRI, but not by CT (Table 1). Bland–Altman plots revealed an even distribution of angles above and below the zero line (Fig. 5). A maximum mean absolute difference of 0.70° ± 3.67° was calculated (Table 1). 95% of the ROM analyses had a difference no larger than 5° and only 1.89% (7/370) showed a difference larger than 10°. Most differences occurred in the analysis of flexion (3/37), followed by extension, abduction, and external rotation (2/37). The analysis of internal rotation combined with other movements (especially flexion) showed only small differences between CT and MRI (3/222) (Table 1). An interval over 5 min between the VIBE and HASTE sequences was not associated with a higher rate of discordant results in ROM analysis between MRI and CT (p = 0.83, Table 4).

The aim of this study was to investigate the agreement between MRI and CT in the diagnostic evaluation of hip morphology using predefined morphological parameters and virtual ROM analysis. It is the first study validating a novel MRI-based software tool for semiautomatic ROM analysis and conventional parameters of hip morphology compared to the gold standard, CT. The software used in this study was previously validated for CT [2]. Other studies already used MRI to analyse hip morphology, but none of them performed ROM analysis in comparison to CT [4, 12, 13].

The most important result of our study is that we found high correlation of radiological parameters and virtual ROM derived from MRI datasets with parameters derived from CT. There was no systematic bias for any of the measures as demonstrated by the Bland Altman plots and the intraclass correlations were 'almost perfect' (ICCs ≥ 0.8) for all radiological measures except the alpha angle at 9 o'clock.

Only a few studies have so far investigated semiautomatic or automatic segmentation of MRI datasets. These studies focused on static parameters and did not perform intermethod correlation. Goronzy et al. investigated manual measurements of acetabular anteversion and coverage of the femoral head and report mean absolute differences not exceeding 0.65° ± 0.86° for inter- and intraobserver as well as intermethod comparison [8]. Chu et al. compared manual and fully automatic segmentation of pelvic CT scans and found mean absolute differences of 2.0° ± 1.5° for acetabular version [14]. Moreover, they found a mean absolute difference of 3.5° ± 2.3° and a range of up to 10° for acetabular coverage. With the semiautomatic software tool investigated here, we found comparable results with a smaller range, suggesting that this type of segmentation is robust and that the results are reliable. Still, manual segmentation seems to be more precise than semiautomatic segmentation.

For ASA a very high intermethod correlation was shown (ICC: 0.98–0.99). The differences in ASA may be explained by the difficulty in differentiating between bone and labrum in MRI data, thereby leading to assumed over coverage. The mean differences of ASA were similar to the manual measurements performed by Goronzy et al. [8]. Exact measurement of the alpha angle can be demanding in both CT and MRI. Yan et al. compared manually measured alpha angles in CT and MRI, reporting a 95% CI up to 10° with an ICC between 0.60 and 0.78 [15]. Furthermore, they showed that agreement between CT and MRI depended on the underlying deformity, observing greater discrepancies for alpha angles in Cam deformities and acetabular parameters in dysplasia. Dessouky et al. report fair to moderate agreement for CT with manual segmentation in both interobserver analysis (ICC: 0.48–0.05) and intraobserver analysis (ICC: 0.49–0.24) [16]. At the same time Xia et al. present a patient collective in which they measured alpha angles in different MRI sequences with a fully automated software, as well as manually, and measured good ICCs (anterosuperior 0.84, anterior 0.92) [13]. The results Xia et al. report for the fully automated approach are slightly inferior to the results we obtained by semiautomatic measurement in the anterosuperior aspect of the femoral neck using CT and MRI (ICC: 0.92–0.97). Agreement for the alpha angle in the posterior part of the femoral head was still good but weaker (ICC: 0.73–0.89). Overall, the alpha angle seems more difficult to measure than other hip parameters such as acetabular version or coverage. Xia et al. found that this problem may be attributable to focal ossifications at the femoral head-neck junction, which may make it difficult to identify the contour of the femoral head, especially in MRI [13]. It may be for this reason that larger differences for the alpha angles in comparison to other pelvic measurements were found with 13 out of 259 cases (5%) exceeding 10∘ difference. Although results showed a good reliability using semiautomatic segmentation and measurements, the software may produce mistakes analysing irregular anatomy.

Although only 7 of the 370 virtual ROM analyses (1.98%) showed larger differences, they need to be discussed critically. Four of these measurements occurred in one patient. The HASTE images of this patient were degraded by motion artefacts distorting the bony outline of the femoral condyles. In this patient, not only the virtual ROM measurements but also femoral torsion measurements showed greater differences between MRI and CT. This case emphasizes the need for proper immobilisation of the legs during MRI and the importance of correlating virtually derived parameters with clinical findings. Not having excluded this case from analysis is a limitation of our study. At the same time, it is a good example of possible sources of error to be encountered when using this method in daily practice.

Regarding femoral torsion, we discovered small inaccuracies of the orientation points on either the femoral condyles or the femoral neck in both, CT and MRI (Fig. 3). Such inaccuracies are especially common for orientation points at the knee condyles due to the greater slice thickness of the HASTE sequence at this level. Botser et al. found higher interobserver reliability for CT than for MRI in manual measurements of femoral torsion. They also report that 77% of the cases had differences larger than 5° [7], whereas in our study only 14% of the cases exceeded this limit. Therefore, our results are superior to previous findings. Conversely, Tomczak et al. reported high intra- and interindividual reliability of femoral torsion measurements for both, CT and MRI [9]. Overall, these findings suggest that careful placement of orientation points by an experienced technician is desirable for both modalities. Additionally, the longer examination time using MRI may lead to leg movement between acquisition of different pulse sequences. In our study, we stabilised patients’ legs during the MRI examinations but VIBE and HASTE sequences were not always acquired consecutively, which is a possible drawback of our study. This can affect ROM analysis in general and parameters such as femoral torsion and increases the risk of motion artefacts. However, in our cohort, we did not find a relationship between longer intervals between the HASTE and VIBE sequences in our patients and outliers. Further, we did not detect a systematic relationship between discrepant torsion measurements and differences between MRI and CT in ROM analysis. Still, we advise to ensure adequate leg fixation and consecutive acquisition of the pulse sequences used for torsion measurement. Therefore, in the course of our study, we developed a special splint to immobilise the feet and thus minimise leg movement during MRI examinations for even more reliable measurements in the future. Overall, the difference in femoral version between CT and MRI was 0.35° ± 3.12° (− 6.3° to 6.5°).

A further limitation of our study is that only pathological angles obtained by ROM simulation using imaging data of patients were available, but no reference values from normal individuals. We do not have detailed information on the interanalyst variation of the technicians who performed manual segmentation. Another limitation of our study is the small number of patients.

However, the mean difference in impingement detected by CT- and MRI-based ROM analysis of − 0.1° ± 2.4° (− 21° to 13°) indicates that MRI-based ROM analysis is sufficiently accurate. There was a 97% match of ROM findings between CT and MRI. In 21.6% of the cases an abnormal ROM was detected.

Omission of CT can reduce total costs; however, MRI is the more expensive modality. The extended MRI protocol used in our study is only 4 min longer than our routine pelvic MRI protocol. However, general contraindications to MRI apply. A diagnostic CT scan of the pelvis exposes the gonads to an ionizing radiation dose of 0.02 Gy, increasing the risk of hereditable diseases by 5% [17].

More studies are needed to further investigate the performance of semiautomatic ROM analysis based on CT and MRI and the potential for fully automated ROM measurement. These studies should include more patients with different types of hip deformities. Moreover, they should address the influence of soft tissue on results. In particular, the integration of the labrum as another stabilising component of the hip will be another important step in the advancement of virtual ROM analysis. MRI-based virtual planning of correction osteotomies and virtual ROM analysis for assessing the outcome of surgery are possible future clinical applications. Another promising application to be investigated is the feasibility of semiautomatic and automatic virtual ROM analysis of MR images obtained in children.

Overall, on most measures, MRI-based semiautomatic analysis of hip morphology and virtual ROM analysis is equivalent to CT. Our findings suggest that, in patients with hip deformities, an MRI alone, with its unique capability to simultaneously image soft tissue pathology (e.g., related to cartilage and the labrum) and bony contours, would suffice in the preoperative analysis of ROM and hip morphology. A separate CT scan with its consequent cost and radiation exposure would not be required and should be avoided, especially when examining younger individuals.