Not All Cams Are Equal: MRI‐Based Statistical Shape Modeling Reveals High Variability in Cam Shape in a Large FAI Cohort

ABSTRACT

Evidence indicates that alpha angle (AA) is inadequate for describing cam morphology, contributing to poor patient selection and clinical outcomes for cam‐type femoroacetabular impingement syndrome (FAI). While techniques such as statistical shape modeling (SSM) have advanced knowledge of 3D femoral morphology, many studies utilize computed tomography, which incurs ionizing radiation, or small sample sizes, limiting clinical implementation and generalization. Here, we conducted SSM using MRI‐derived 3D femur models from a large cohort of FAI patients to explore: (1) What proximal femur features drive shape variability? (2) What variability exists within the area of the cam deformity? (3) Are shape modes correlated to AA, age, sex, and BMI? 103 FAI patients were enrolled (µ ± σ): Age (29.9 ± 12.1 Years), BMI (24.6 ± 4.8), AA (60.3° ± 9.8), Female/Male (75/28). Shape modes were determined through principal component analysis. Relationships between shape modes and AA, age, and BMI were evaluated through linear regression. Sex differences were analyzed using linear discriminant analysis. Thirteen modes explained 91.0% of femoral shape variance, with the first six modes explaining 82.2%, four of which were associated with the cam deformity region. No relationship was found between any mode versus AA, BMI, or age. Sex‐based shape differences were primarily captured by the first mode.

Clinical Significance

The proximal femur in cam‐type FAI patients demonstrated substantial regional shape variability, with cam‐related morphology distributed across multiple independent modes rather than a single uniform deformity pattern. Sex was a major driver of proximal femoral shape variation. These findings may inform morphology‐based classification, outcome analysis, and patient‐specific surgical planning.

1. Introduction

Cam‐type femoroacetabular impingement (FAI), characterized by an abnormal bony growth along the femoral head‐neck junction [1, 2], is a common cause of hip pain in young individuals and active adults [3, 4] and is a known precursor for the development of hip osteoarthritis (OA) [3, 5, 6]. It is estimated that the incidence of FAI in the general population is 54.4 per 100,000 person‐years, with females having a significantly higher incidence compared to males [7]. The mechanism through which cam FAI leads to hip OA is through mechanical joint impingement caused by abnormal proximal femoral shape, described as a lack of offset at the head‐neck junction [8]. Repeated combinations of hip movements and positions such as hip flexion, internal rotation, and adduction in combination with this abnormal proximal femoral shape [9] lead to hip impingement [10, 11, 12, 13]. This impingement results in contact with the chondrolabral junction [1, 2, 14, 15], which can lead to labral tears, cartilage damage [16], and eventually OA [3, 5, 6]. Clinically, cam‐type FAI typically presents as symptoms of hip pain [1, 17], reduced range of motion [18], and loss of functional movement [19, 20].

Hip arthroscopy with osteochondroplasty, in which the surgeon reshapes the proximal femur by removing the cam morphology, has emerged as a primary treatment for cam‐type FAI after failure of conservative management [21, 22]. It is well recognized that both under‐ and over‐resection of FAI is a primary risk factor for poor outcomes and need for revision surgery [23, 24, 25, 26, 27, 28]. Accordingly, being able to accurately quantify and understand bony shape is central to understanding disease severity and optimizing outcomes for arthroscopic FAI management [9, 14, 15, 29]. A standard set of hip planar radiographs, including assessment of the femoral alpha angle [30], is currently the gold standard for clinical diagnosis. The alpha angle measurement is made on a 2D planar radiograph by calculating the angle between the axis of the femoral neck and a line that connects the center of the femoral head with a point designating the beginning of asphericity on the head‐neck contour [30]. The alpha angle remains the most widely used metric for quantifying femoral head asphericity in cam‐type FAI, though its accuracy, repeatability, and correlation with intraoperative findings have been questioned [31, 32]. A threshold of > 55° has traditionally defined pathologic morphology, though alternative cutoffs have since been proposed, reflecting ongoing lack of consensus [30, 33, 34, 35]. In clinical practice, AP pelvis and cross‐table lateral radiographs serve as practical first‐line modalities given their accessibility and efficiency, and the Dunn view at 45° or 90° of hip flexion improves sensitivity by yielding higher alpha angle measurements [36]. Nevertheless, all 2D projections risk underestimating an inherently 3D deformity, particularly at the anterosuperior femoral head‐neck junction where maximal cam extent is most commonly located [7, 37, 38, 39].

While the alpha angle remains clinically useful because it is easy to understand, widely known, and easily incorporated into routine imaging workflows, it reduces the complex 3D morphology of the cam deformity to a single 2D planar measurement. This is an important limitation, as the size, location, and spatial extent of the deformity can vary substantially around the femoral head‐neck junction [38, 39, 40, 41, 42]. 3D imaging, including MRI and CT, provides a more complete picture of proximal femoral morphology and may be particularly useful for understanding cam location, deformity extent, and patient‐specific regions at risk for under‐ or over‐resection during osteochondroplasty. MRI‐derived 3D models are especially attractive in this context because they avoid ionizing radiation and may allow osseous morphology to be evaluated alongside cartilage, labral, and other soft‐tissue features from the same imaging session. However, despite the potential value of cam morphology assessments using MRI, studies that comprehensively characterize proximal femoral shape variability in large cohorts of patients with cam‐type FAI remain limited.

Radial imaging has improved the assessment of cam morphology by allowing alpha angle measurements to be obtained across multiple planes around the femoral head‐neck junction [30, 40, 41, 42]. Several studies have shown that radial alpha angles can identify deformities that may be underestimated or missed on conventional radiographic views [40, 41]. However, radial alpha angle measurements still reduce morphology to a series of planar angular measurements and do not fully describe the 3D size, shape, or spatial distribution of the cam deformity. Therefore, although radial imaging remains useful in clinical practice, surface‐based 3D assessments may provide complementary information by characterizing the full proximal femoral morphology and the regional variability of cam‐related shape features.

Previously, our group developed a non‐invasive shape‐fitting method to virtually predict cam deformity volume in FAI patients, for the purposes of aiding surgical planning [43]. While this technique can be used to reveal quantitative information about the cam geometry at a patient‐specific level pre‐operatively, we sought to enhance this prior work by using population‐based statistical shape modeling (SSM), a technique that has increasingly been implemented in studies on FAI [44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54]. SSM is a population‐based technique that generates a statistical representation of a cohort of 2D or 3D geometries, such as the shape of the proximal femur, that can be used to quantify shape variations within cohorts [55, 56]. As cam‐type FAI is a disorder that is directly related to abnormalities of the underlying femoral morphology, this makes SSM an optimal tool for quantifying the cam deformity. SSM has previously been applied to evaluate proximal femoral shape in patients with FAI to determine the spatial distribution of the cam deformity, identify association with radiographic femoral measures such as alpha angle and head‐neck offset, and also to determine mean shape differences between cam FAI patients versus controls [9, 45, 46, 47, 48, 49, 50, 51, 52, 54, 57, 58]. Furthermore, existing literature discusses sex‐specific variations in cam shape, location, and volume [51]. However, a large majority of studies are limited by small patient cohorts (< 100 subjects), as well as utilizing computed tomography (CT) scans [45, 46, 47, 48, 49, 52, 54, 57, 58], which incurs ionizing radiation to subjects and precludes obtaining a follow‐up scan. The purpose of this study was therefore to apply SSM to a large cohort of MRI‐derived 3D proximal femur models from patients with cam‐type FAI in order to characterize dominant patterns of proximal femoral shape variation, define regional variability within the typical cam deformity zone, and determine whether shape modes were associated with alpha angle, age, sex, or BMI [59].

2. Methods

2.1. Patient Recruitment

For this institutional review board‐approved, retrospective level 3 study, 103 total patients with FAI were enrolled from a single surgeon's high‐volume hip preservation clinical practice (2017–2020). Inclusion criteria for this study included: clinical diagnosis of cam‐type FAIS based on clinical examination and radiographic imaging, and underwent Flash‐Dixon 1.5 T magnetic resonance imaging (MRI). Exclusion criteria for the study included: radiographic evidence of hip osteoarthritis defined as Tönnis Grade > 1, and history of a developmental hip disorder such as Legg‐Calvé‐Perthes disease or slipped capital femoral epiphysis. For our assessment of anatomic features driving shape variability of the proximal femur, we used MRI scans from all patients recruited, resulting in a sample size of N = 103.

2.2. Magnetic Resonance Imaging and Segmentation

Deidentified 1.5 T Flash‐Dixon MRI scans were obtained (voxel resolution 0.9 × 0.9 × 0.9 mm isotropic), as previously described [60, 61], for manual segmentation in 3D Slicer (Figure 1). Segmentations were generated using a manually supervised workflow [60]. Thresholding was used to assist initial identification of the proximal femur, and then segmentations were manually adjusted. Any remaining segmentation island artifacts were removed prior to smoothing using 3D Slicer's native Islands tool (Keep Largest Island function).

Figure 1.

Representative MRI slices (top row) and corresponding segmentations (bottom row) in the coronal (A), sagittal (B), and axial (C) planes. Segmented regions are indicated by green fill confined within cortical boundaries.

A combination of closing, opening, and median smoothing operations was applied to the binary segmentation masks in 3D Slicer to reduce voxel‐level artifacts, fill small holes, and remove small non‐anatomic irregularities prior to surface export. The following voxel‐level smoothing steps were used: (1) “Closing (fill holes),” (2) “Opening (remove extrusions)” at a kernel size of 5.00 mm, and (3) “Median smoothing” at a kernel size of 3.00 mm. After these segmentation‐specific smoothing steps, segmentation accuracy was manually verified slice‐by‐slice across the axial, coronal, and sagittal planes to confirm complete capture of the proximal femur bone within cortical boundaries and to identify any residual segmentation artifacts prior to surface export. All MRI scans were successfully segmented; no subjects were excluded on the basis of poor image quality or segmentation failure.

2.3. SSM

Segmented femoral 3D bone models were imported as. STL files into an open‐source SSM software [62] (ShapeWorks). To ensure consistency between all femurs, constraints were applied to each 3D model by defining a plane perpendicular to the axis of the long bone of the femur positioned directly above the lesser trochanter (Figure 2). Preprocessing for SSM was then done by first reflecting the left hips along the sagittal plane so all hips were directly comparable regardless of side. Then all models were rigidly aligned using an iterative closest point algorithm. Following reflection and rigid alignment, all exported STL surfaces underwent volume‐preserving Laplacian smoothing as a mesh‐level grooming step prior to the particle correspondence step. This step was distinct from the earlier voxel‐level smoothing performed during segmentation and was applied to reduce surface artifacts from MRI voxelization, segmentation boundaries, and mesh generation while preserving the underlying proximal femoral morphology. The number of correspondence particles was determined by evaluating SSM quality metrics (i.e., compactness, generalization, and specificity) across iteratively doubled particle configurations ranging from 256 to 4096, following established ShapeWorks optimization protocols [62]. Based on these quality metrics, 4096 particles were selected, consistent with previously reported proximal femur SSMs [52]. A total of 4096 correspondence particles with a relative weighting of 50 between the correspondence and surface sampling entropy terms [63] were therefore placed and optimized on each subject's proximal femur surface in ShapeWorks [62]. Unlike landmark‐based or template‐driven SSM approaches, the particle‐based method described by Cates et al. optimizes correspondence across the entire shape ensemble simultaneously, producing a more compact and geometrically accurate statistical shape model [62]. To reduce the effect of both femoral size and alignment, a Procrustes transformation was used in the analysis.

Figure 2.

Representative 3D model, demonstrating constraint placement and particle distribution.

2.4. Statistical Analysis

The mean shape of the femurs was then calculated based on the mean particle configuration. An assessment using principal component analysis (PCA) to determine geometrical variations in shape across our cohort of cam FAI patients was performed. Here, principal components (PCs), or shape modes, and their respective explained variance were determined. Shape variation for each mode was reviewed using animations of each principal component across ±2 standard deviations from the mean shape and static surface displacement maps showing regional surface differences. Animations were used to assess the overall pattern of shape change, while the surface maps were used to identify the anatomic regions with the largest displacements and to estimate approximate displacement ranges. Regional shape variation was categorized by the magnitude of surface displacement from the mean shape: high (> 1.5 mm), moderate (1.0–1.5 mm), and low (< 1 mm). As an exploratory analysis, associations between shape modes and alpha angle, age, and BMI were evaluated to determine whether common clinical and demographic variables explained dominant patterns of 3D proximal femoral shape variation. Alpha angle was included because it remains the most widely used clinical measure of cam morphology, while age and BMI were included to assess whether dominant shape modes were associated with basic patient characteristics. Linear discriminant analysis (LDA) was conducted to evaluate sex‐based differences in the SSM. Shape mode scores from the modes explaining ≥ 90% of total variance were used as candidate input features. Minimum‐redundancy maximum‐relevance feature selection (MRMR) [64] was used to identify the subset of shape modes contributing most strongly to sex‐based separation. Classification performance was summarized using the area under the receiver operating characteristic curve (AUC). Statistical significance was evaluated using permutation testing, in which sex labels were randomly shuffled, and the LDA workflow was repeated to generate a null distribution of Mahalanobis distances between groups (α = 0.05). As a negative‐control quality check, the same LDA workflow was also used to test whether the model separated subjects by original hip laterality after left hips were reflected and all models were aligned.

3. Results

The patient cohort enrolled in this study had the following characteristics (mean ± standard deviation): Age (29.9 ± 12.1 Years), BMI (24.6 ± 4.8), AA (60.3° ± 9.8). This cohort included 28 males and 75 females, for a total of N = 103 subjects. From the SSM model, the first 6 shape modes (principal components) had a cumulative explained variance of 82.2%, and the first 13 shape modes had a cumulative explained variance of 91.0% (Figure 3). Descriptions of each mode and their explained variance in the model are provided in Table 1. Visualizations of modes 1–6 can be seen in Figure 4.

Figure 3.

SSM Metrics of compactness. The vertical dotted line is marked at 13 Modes, which was found to have a cumulative explained variance of > 90% (horizontal dotted line).

Table 1.

SSM Shape Modes, including explained variance for each mode, and description. The cumulative explained variance of the first 13 modes of the model is 91.0%.

Shape mode	Explained variance	Cumulative variance	Description
1	41.3%	41.3%	High variability exists about the circumferential ring at the femoral head‐neck junction (2.7 to –2.6 mm) and medial femoral neck (2.6 to –2.6 mm). There is also a focal area of high variability at the distal anteromedial femoral neck (2.5 to –2.5 mm). Moderate variability also exists within the posteromedial femoral neck (2.0 to −1.8 mm), albeit not to the spatial extent compared to the anteromedial side, as well as the superomedial femoral head (1.0 to –1.1 mm) and superolateral greater trochanter (1.4 to –1.5 mm).
2	21.1%	62.4%	The highest variability begins at the superior greater trochanter, extending proximally to the base of the femoral neck (3.2 to –3.1 mm). High variability is also appreciated at the intertrochanteric crest (2.5 to –2.0 mm). Mild to moderate variability is noted on the medial surface of the femoral head (1.1 to –0.9 mm), with mild variation being observed on the lateral greater trochanter (0.6 to –0.6 mm).
3	9.2%	71.6%	High variability is observed in three locations within Shape Mode 3: the intertrochanteric fossa (2.0 to –2.0 mm); the posterolateral greater trochanter (1.8 to –1.7 mm); the distal anteromedial femoral neck with extension superiorly along the intertrochanteric line (1.6 to –1.8 mm). There was moderate variability (1.0 to –1.2 mm) proximal to this region, which extends superiorly along the intertrochanteric line. Moderate variability is observed at the circumferential ring at the femoral head‐neck junction (1.5 to –1.5 mm).
4	5.1%	76.6%	Moderate variability is displayed at the anterior surface of the femur distal to the intertrochanteric line (1.8 to –1.7 mm) and the posteromedial greater trochanter with extension inferiorly along the lateral border of the intertrochanteric crest (1.6 to –1.8 mm). Moderate variation is observed at the superior femoral neck (1.4 to –1.4 mm), with mild variation noted at the superior posterior aspect of the greater trochanter (0.8 to –1.0 mm).
5	2.9%	79.6%	High variability is observed at the anterosuperior femoral neck with extension inferiorly along the medial border of the intertrochanteric crest (1.5 to –1.6 mm), while moderate variability is seen at the superior aspect of the greater trochanter (1.4 to –1.2 mm). Mild variability is appreciated at the superior femoral head‐neck junction (0.7 to –0.8 mm), the inferior femoral head‐neck junction with inferior extension along the medial femoral neck (0.8 to –0.8 mm), within the fovea capitis (0.6 to –0.7 mm), and at the intertrochanteric crest (0.8 to –0.8 mm).
6	2.6%	82.2%	Moderate variation is appreciated at four locations: the anterior head‐neck junction (1.1 to –1.2 mm), posterior superior aspect of the greater trochanter (1.0 to –1.0 mm), the lateral portion of the superior femoral neck (1.3 to –1.4 mm), and at the junction of the intertrochanteric crest and posterior femoral neck with extension superiorly into the intertrochanteric fossa (1.3 to –1.0 mm). Mild variation is also noted at the lateral greater trochanter (0.7 to –0.7 mm).
7	2.0%	84.1%	Moderate variability appears at the superomedial greater trochanter (1.5 to –1.5 mm), with mild variability being observed along the medial femoral neck (0.8 to –0.8 mm), posterior greater trochanter (0.5 to –0.5 mm), and the junction of the intertrochanteric crest and femoral neck, which extends superiorly just into the intertrochanteric fossa (0.8 to –0.7 mm).
8	1.8%	85.9%	The intertrochanteric fossa and posteromedial femoral neck demonstrate moderate variability (1.4 to –1.4 mm; 1.0 to –1.0 mm). Mild variation is observed at the superior aspect of the greater trochanter (0.9 to –0.7 mm); the anterior surface of the femoral neck bordered inferiorly by the intertrochanteric line and superiorly by the anterior femoral head‐neck junction (0.6 to –0.6 mm); and the anterior inferior greater trochanter, just lateral to the intertrochanteric line (–0.5 to 0.5 mm).
9	1.2%	87.1%	Mild variation was observed at several locations: the anterosuperior femoral neck (1.1 to –0.9 mm), the superior posterior facet of the greater trochanter (0.6 to –0.6 mm), the intertrochanteric line (0.5 to –0.5 mm), a focal area within the intertrochanteric fossa (1.0 to –0.9 mm), and along the intertrochanteric crest (0.7 to –0.7 mm).
10	1.2%	88.2%	Moderate variation is appreciated at the anterior femoral head‐neck junction (1.1 to –1.1 mm) and the anterior inferior greater trochanter, just lateral to the intertrochanteric line (1.0 to –1.0 mm). Mild variability was observed at several locations, including the intertrochanteric line (0.9 to –0.9 mm); inferior edge of the fovea capitis (0.8 to –0.8 mm); inferior aspect of the circumferential ring at the head‐neck junction (0.7 to –0.7 mm); superior posterior facet of the greater trochanter (0.7 to –0.7 mm); and the superior section of the femoral neck (0.5 to –0.5 mm).
11	1.0%	89.2%	Mild variation is noted at the superior femoral neck (0.6 to –0.7 mm), intertrochanteric fossa (0.6 to –0.6 mm), superior posterior facet of the greater trochanter (0.7 to –0.7 mm), and the inferior head‐neck junction with anterior circumferential extension (0.8 to –0.8 mm).
12	1.0%	90.2%	Mild variation is evident at several locations: superior posterior facet extending inferiorly into the intertrochanteric fossa (0.5 to –0.5 mm), superior section of the intertrochanteric line (–0.6 to 0.6 mm), intertrochanteric crest (0.5 to –0.6 mm), anterior section of the head‐neck junction (–0.8 to 0.8 mm), anterior femoral neck (0.5 to –0.6 mm), and femoral head just superior to fovea capitis (–0.8 to 0.8 mm).
13	0.8%	91.0%	Multiple areas of mild variability were observed, including the intertrochanteric line (0.4 to –0.4 mm), medial border of the intertrochanteric crest (0.8 to –0.7 mm), superior anterolateral aspect of the greater trochanter (0.4 to –0.5 mm), anterosuperior head‐neck junction (0.5 to –0.5 mm), medial femoral neck (0.6 to –0.6 mm), and circumferentially about the fovea capitis (0.4 to –0.5 mm).

Figure 4.

Visualization of Shape Modes 1–6 for the proximal femur in three views, which together describe 82.2% of the explained variance in the model. Views are shown as follows: anterior view, top row; superior view, middle row; posterior view, bottom row. For each mode, columns show shape variation from −2 SD to +2 SD relative to the mean shape. The color bar indicates surface displacement from the mean shape in millimeters. Semi‐quantitative descriptions of the regional changes associated with each mode are provided in Table 1.

Using linear regression, no significant relationship was found between any shape mode and alpha angle, BMI, or age. LDA with MRMR feature selection identified a statistically significant separation between male and female subjects (p = 0.0005; Figure 5A). After MRMR feature selection, the resulting LDA model retained only Shape Mode 1 and demonstrated strong classification performance, with an AUC of 0.973. This indicates that sex‐based shape differences in this cohort were primarily captured by a single shape mode. Clear separation between male and female subjects can be seen in Figure 5B, which plots individual subjects by Shape Mode 1 and Shape Mode 2. A visualization of the average proximal femur shape for males and females, alongside the ±2 standard deviation shapes associated with Mode 1, is shown in Figure 6. Because Procrustes scaling was used, the observed male‐female differences were not simply driven by overall femur size, but instead reflect differences in overall proximal femoral shape. No significant separation was observed between left and right hips, supporting that the sex‐based separation was not simply an artifact of the LDA workflow.

Figure 5.

Sex‐based separation of proximal femoral shape. (A) Linear discriminant analysis (LDA) of male and female subjects after MRMR feature selection demonstrated significant separation between groups (p = 0.0005). The density curves show the distribution of LDA scores for female and male subjects, with individual subject scores shown along the x‐axis. (B) Scatter plot of individual subjects by Shape Mode 1 and Shape Mode 2. Female subjects are shown as orange circles and male subjects as blue squares. Separation between groups occurred primarily along Shape Mode 1, indicating that sex‐based differences in proximal femoral morphology were largely captured by the dominant mode of shape variation.

Figure 6.

(Left) Visualization of the average proximal femur shape for females versus males. The color bar represents the difference from the average shape in each group (female or male) from the overall average mean (i.e., mean shape of entire cohort – mean shape of each sex, respectively). (Right) The ±2 standard deviations from the mean shape of the proximal femur associated with Mode 1 are shown to demonstrate similarities to the average shapes for females and males (Mode 1 femurs are also shown in Figure 4). Differences were most apparent around the femoral head‐neck junction, medial/posteromedial femoral neck, distal anteromedial femoral neck, and superolateral greater trochanter, while other regions showed more similar morphology between groups.

4. Discussion

We developed an SSM workflow to assess variability in proximal femoral shape in a large cohort of patients with FAI. This study applies established SSM methods to one of the largest MRI‐derived proximal femur cohorts reported in cam‐type FAI. Our SSM model demonstrated that ~82.2% of the cumulative variance was explained by 6 shape modes, and 91.0% of the cumulative variance was explained by 13 shape modes. This result aligns closely with a previous study by Harris et al. from 2013 [54], which identified 12 shape modes that explained ~90% of the variance in their model. This agreement with prior SSM work is an important finding, as it supports that SSM captures reproducible patterns of proximal femoral shape variation across different cohorts and imaging modalities. Importantly, in our model, of shape modes within the first 82.2% of cumulative variability, shape modes 1, 4, 5, and 6 were all shown to be directly related to the area of the cam deformity on the head‐neck junction, demonstrating that a high degree of variability in the shape of the cam deformity exists within this large cohort of cam FAI patients. This is an important finding, demonstrating that not all cams are structurally the same.

While SSM methodologies have increased in popularity over the past decade, this study is unique in several ways, including that a large cohort of over 100 patients was included in the analysis, and that MRI was used to obtain 3D models, the latter of which has only recently been reported [9, 50, 51]. The field of hip preservation has also reported increased use of MRI [65, 66] as a part of clinical diagnosis for abnormal hip morphology, as it can generate high‐quality and high‐resolution images that can be segmented into accurate 3D models, providing a far superior representation of the joint's morphology than a traditionally used 2D radiographic scan [61]. In addition, the use of MRI is significantly advantageous compared to other common image modalities used in SSM studies, such as radiography and CT imaging, in that it can be conducted without exposing patients and medical staff to ionizing radiation, which is consistent with ALARA principles for minimizing radiation exposure [67]. Thus, the utility of these imaging modalities is limited for high‐throughput studies with large cohorts of subjects compared to the use of MRI, which provides a safer imaging modality with reasonable resolution for SSM studies.

4.1. Variability of the Femoral Head and Head‐Neck Junction

Our SSM model indicated that there is substantial variability associated with the femoral head. In Shape Mode 1, which explained the largest amount of variability, a primary feature captured by this mode was the aspect ratio of the femoral head, with one extreme of 2 standard deviations depicting a more conchoid‐shaped femoral head [68], and the other depicting a more spherical‐shaped femoral head. This change across ±2 standard deviations of Mode 1 was associated with minimal to no change in the length of the femoral neck. This is an intriguing finding, as a shorter femoral neck with a large cam deformity may have a higher likelihood of impingement during movement [69, 70]. In addition, the variability observed on the anterior side of the proximal femur suggests that the cam deformity itself may have variability that is correlated with the circumference around the femoral head, demonstrating that the cam deformity's shape may be associated with external structures on the proximal femur. More interestingly, Shape Mode 1 was also strongly associated with sex‐based differences, as discussed below.

In addition to Mode 1, Modes 6, 8, and 10 demonstrated variability associated with the angle of the femoral head relative to the femoral neck, which can contribute to femoral version. This is consistent with prior work suggesting that femoral version may influence cam‐type FAI biomechanics, as femurs with anteversion may have a higher likelihood of experiencing impingement. However, it is important to note that the presented SSM cannot describe how the version in the context of the position of the femoral condyles contributes to the variance observed in the model, as these structures are not included in the study. In future work, we intend to assess the entire femur to more robustly assess the role of femoral version in the context of cam FAI.

4.2. Variability of the Greater Trochanter

Shape Modes 2 and 3, which explained 21.1% and 9.2% of the variance in the model, respectively, were associated with variations in the shape of the greater trochanter and adjacent proximal femur. These findings are consistent with prior SSM studies of cam FAI and related hip morphology cohorts. Harris et al. [52] reported that early modes of variation in cam FAI included trochanteric height and femoral neck width, while Harris et al. [54] found that greater trochanter height and femoral offset were shared sources of variability between DDH and cam FAI, with greater trochanter contour more specifically contributing to variation in the cam FAI group. In the present study, Mode 2 was primarily associated with variability beginning at the superior greater trochanter and extending proximally to the base of the femoral neck, as well as high variability along the intertrochanteric crest. Mild to moderate variability was also observed on the medial surface of the femoral head, with mild variation on the lateral greater trochanter. In contrast, Mode 3 demonstrated high variability at the intertrochanteric fossa, posterolateral greater trochanter, and distal anteromedial femoral neck, with variation extending superiorly along the intertrochanteric line. Moderate variability was also observed at the circumferential ring of the femoral head‐neck junction.

Similar to the greater trochanter height, contour, and femoral offset findings described by Harris et al. 2024 and Harris et al. 2013, Modes 2 and 3 in the present study appear to reflect broader proximal femoral shape variation rather than cam‐specific morphology alone [52, 54]. While neither Mode 2 nor Mode 3 was primarily associated with the anterosuperior aspect of the proximal femur, where the cam deformity is typically found, these modes affected regions of the greater trochanter and intertrochanteric region where major muscle attachments are located. The association between proximal femur bone shape and muscle insertion positions has implications for the biomechanics of the joint, and therefore may contribute to hip impingement. Furthermore, muscle activation, especially during sporting activity, may also affect skeletal development in adolescents [71], which could indirectly impact the development of proximal femur shape when the bone growth plates are not closed. Further research is needed to confirm whether this postulation is valid and may be an interesting focus of future research.

Shape Mode 4 was associated with the posterior aspect of the superior posterior facet of the greater trochanter, as well as the superior aspect of the femoral neck, the latter of which was located at the 12 o'clock position. Mode 4 also indicated localized areas with variance in shape on the femoral neck, in a location where the cam deformity can be observed.

4.3. Variability Within the Cam Zone (12 O'Clock to 3 O'Clock Radial Positions on the Proximal Femur)

Several shape modes were identified where the explained variance was associated with cam‐specific features. In addition to Modes 1 and 4 already discussed, Modes 5–6 had shape variations within the cam zone, where both had isolated patterns of variance that were distinct. Mode 5's explained variance corresponded spatially to more surface area on the proximal femur, with variations in the anterior, superior, and lateral aspects of the femoral head‐neck junction. Mode 6 also demonstrated shape variations within the cam zone, but was primarily associated with changes in shape at the head‐neck junction toward the femoral head. Thus, Modes 5 and 6 represent different variances associated with the cam deformity. In addition, Modes 8–12 also demonstrated areas of variance on the proximal femur associated with where the cam deformity is located.

4.4. Considerations of Higher Shape Modes and the Contribution of Model Noise

While SSM provides an unbiased probabilistic method to assess shape variations, there are still limitations associated with this method. While shape Modes 7–13 had a cumulative explained variance of 8.8%, each of these shape modes only contributed at most 2.0% of the explained variance individually. Therefore, these data must be considered carefully to determine whether the variance is driven by actual physical features or by noise. Model noise can be caused by several factors, including poor segmentation accuracy and MRI resolution (slice thickness). Furthermore, in these higher shape modes, the associated changes in shape were mostly less than 1 mm.

4.5. SSM Implications on Diagnosis and Surgical Intervention

A takeaway of this study, consistent with prior SSM and 3D imaging studies, is that substantial variability exists in proximal femoral shape among patients with cam‐type FAI [50, 52, 54, 72]. Given that several modes of variation were involved with the area where the cam deformity is typically located (12–3 o'clock regions of the femoral head and neck), we postulate that cam‐type FAI is a highly heterogeneous condition morphologically. The data presented suggest that more robust measures of femoral structural morphology are needed to capture this variability and emphasize the potential utility of 3D imaging modalities. More recently, the Omega angle has been proposed [39], which takes into account additional metrics of proximal femur shape. However, this measure remains less commonly used and/or reported compared to the alpha angle.

These findings also relate to prior work using SSM to inform cam resection. Atkins et al. demonstrated that correspondence‐based shape modeling can be used to quantify cortical bone thickness in patients with cam FAI and later used SSM to evaluate whether simulated removal of subchondral cortical bone in the cam region restored proximal femoral anatomy toward that of screened controls [46, 47]. In that study, removal of subchondral cortical bone reduced the mean shape deviation between cam and control femurs to within approximately 1 mm, supporting the subchondral cortical‐cancellous bone margin as a potential visual guide for resection depth [47]. In contrast, the present study does not define a resection depth or postoperative target shape. Rather, our results identify multiple regions of cam‐related variability, including the circumferential head‐neck junction, distal anteromedial femoral neck, superior femoral neck, and anterior head‐neck junction, that may need to be considered in future studies linking preoperative morphology, resected volume, postoperative shape, and clinical outcomes.

Therefore, while the present work does not provide direct surgical resection guidelines, it supports the need for patient‐specific 3D characterization of cam morphology. This may be particularly important because current osteochondroplasty planning does not routinely account for the regional variability in cam morphology observed across this cohort, is highly surgeon‐dependent, and lacks objective intraoperative guidance. This variability in resection technique may contribute to inadequate cam correction, and in turn to the need for revision arthroscopy or early conversion to total hip arthroplasty [37].

4.6. Sex as a Driver of Proximal Femoral Shape Variability

While the current study is broadly consistent with prior literature on proximal femoral shape variability [52, 53, 54], it further adds to the existing body of work by showing that sex was the primary driver of shape variability in an unsegregated cam‐type FAI cohort. This separation emerged post hoc, without a priori grouping by sex during SSM and after accounting for size, suggesting that sex corresponds strongly to one of the dominant shape patterns identified in the full cohort.

These findings are consistent with Braun et al., who used 3D SSM to evaluate proximal femoral morphology in a cohort of athletes and identified sex‐based differences involving the femoral head, head‐neck junction, femoral offset, and greater trochanter region [53]. Similarly, in the present study, Shape Mode 1 separated male and female patients and included variation at the circumferential femoral head‐neck junction, medial and posteromedial femoral neck, distal anteromedial femoral neck, and superolateral greater trochanter. Because Procrustes scaling was used, these differences were not driven simply by overall femur size, but instead reflect differences in proximal femoral shape.

Our findings also align with prior imaging‐based studies of sex differences in cam morphology. Yanke et al. used 3D CT‐based quantification and reported that male cam deformities had greater height and volume than female cam deformities, as well as a greater total clockface span of deformity [72]. The present SSM‐derived findings support the broader concept that male and female cam‐type FAI morphology differs in 3D, but also suggest that these differences are not limited to cam size alone. Instead, sex‐associated variation was observed across the head‐neck junction, medial and posteromedial femoral neck, distal anteromedial femoral neck, and greater trochanter region, consistent with SSM Mode 1 in our study.

These results also provide context for the findings of Bugeja et al., who reported sex‐specific differences in cam morphology location using 3D MRI‐based analyses [51]. In the present study, the specific finding most consistent with Bugeja et al. was the distal anteromedial femoral neck variation observed in Shape Mode 1, which was also the mode most strongly associated with sex. This comparison should be interpreted cautiously because our study did not quantify cam location using the same approach as Bugeja et al.; however, the overlap supports the broader conclusion that male and female patients with cam‐type FAI may differ not only in cam size, but also in the regional distribution of proximal femoral shape features.

Importantly, the observed shape differences between males and females spanned regions both within and outside the typical cam deformity region, suggesting that SSM captures not only deformity‐related variation but also broader differences in male and female proximal femoral morphology. These findings may have implications for future surgical planning studies, as osteochondroplasty aims to restore proximal femoral shape while avoiding both under‐ and over‐resection. The present study does not define sex‐specific surgical targets, but it suggests that a single idealized proximal femoral shape may not fully represent the morphology of both male and female patients with cam‐type FAI. Future work incorporating pre‐ and postoperative 3D models, resected cam volume, and clinical outcomes will be needed to determine whether sex‐associated shape differences are relevant to postoperative morphology targets and clinical outcomes.

4.7. Limitations of the Study

While this study presents one of the largest cohorts of FAI patients to be investigated using SSM, there are some limitations that are worth noting. First, this study was retrospectively conducted, using an existing database collected from a single surgeon. Differences in FAI diagnosis and patient selection between practices could contribute to the variability we did not observe. In addition, this study assesses a large cohort of FAI patients, but does not include comparisons to asymptomatic controls. This was in part due to the focus of the study, which was to identify the extent of variability within femoral morphology at the head‐neck junction, as this is a primary target for osteochondroplasty treatment. However, because of a lack of a control group, it is challenging to distinguish between variability that is inherent within a healthy population versus variability that contributes to disease severity. Future work will include asymptomatic controls to address this, in order to determine if arthroscopic surgery corrects the FAI hip shape, such that postoperative hip shape is representative of the asymptomatic population. This would provide crucial information toward improving surgical resection guidelines to avoid under‐ or over‐resection. Another limitation is that we do not consider the acetabulum. However, as osteochondroplasty is performed on the femur only, the results of this study still provide significant and valuable information about the importance of considering the 3D shape of the femur rather than a 2D metric such as the alpha angle.

Another limitation is that the MRI‐derived surfaces were generated from scans with 0.9 mm isotropic voxel resolution. Although all segmentations were manually reviewed slice‐by‐slice and all surfaces underwent preprocessing before SSM, the effects of MRI resolution and segmentation uncertainty were not directly quantified in this study. As a result, smaller surface differences near the voxel resolution, particularly in higher‐order modes, should be interpreted cautiously. Larger regional differences, especially those exceeding approximately 1.5 mm, are less likely to reflect voxel‐level uncertainty alone, but future studies using repeat segmentation, inter‐rater analysis, or paired CT/MRI surface comparisons would be useful to quantify the effect of imaging resolution and modality on SSM results.

4.8. Research Translatability, Future Directions, and Clinical Significance

The present study demonstrates that cam‐type FAI morphology is not a single uniform head‐neck junction deformity, but a heterogeneous 3D shape phenotype involving multiple regional patterns of proximal femoral variation. This highlights the broader utility of SSM for characterizing proximal femoral shape variation and osseous deformity in ways that are difficult to capture on plain radiographs alone. This is particularly relevant given the previously described positive relationship between cam size and labral pathology [31]. Shibata et al. further demonstrated that hip morphology directly influences the pattern of labral and articular cartilage damage [73], underscoring that accurate, patient‐specific morphologic characterization is not merely informative but critical to surgical decision‐making, given that failure to adequately address bony deformity remains the primary driver of poor outcomes and revision surgery in FAI [74, 75, 76, 77]. Consistent with prior imaging and shape‐analysis literature calling for improved assessment beyond standard radiographic and 3D asphericity measurements [31, 34, 38, 41, 78, 79], our findings support future use of SSM as a practical and potentially useful complement to existing imaging‐based workflows.

The next steps for this work will include several pathways, including (1) incorporating assessments of changes in shape before and after osteochondroplasty to assess the resected volume, and correlation of this volume to clinical outcomes (2) assessments of the acetabulum shape, both in isolation and in combination with the femur shape, as cam‐type FAI is not a one‐bone issue, and consideration of the other side of the impingement is also important and (3) investigations on the relationship between hip bone morphology and functional joint biomechanics using motion analysis and/or computational modeling to better understand how different activities, motions, and sports interact with a patient's bony anatomy to lead to impingement. Finally, the SSM models in this study will continue to be optimized with an expanded cohort, with the goal of building a model that is increasingly representative of the patient population. As our cohort grows, we intend to make our SSM model freely available (open‐source) to the orthopedic and sports medicine communities as a valuable resource for studying cam‐type FAI.

The results of this and future studies will serve as a foundational first step toward building a novel predictive platform based on femoral shape variation, motion analysis, and hip joint contact mechanics. Because SSM is built through a quantitative assessment of individual patients’ shape features, it presents opportunities to build classifiers based on joint shape. Thus, a future direction of the presented work will be to build classifiers using unsupervised clustering and regression techniques to determine if there are distinct patterns of shape that are associated with specific movement patterns or contact. Identified clusters of patients may in turn be associated with patient demographics, outcomes, and subsequent development of osteoarthritis [80]. By elucidating such patient clusters and distinct patterns, this research may provide insight that can lead to improved guidelines for patient selection, surgical planning, and clinical decision‐making.

Author Contributions

Catherine Yuh: [1] substantial contributions to research design, the acquisition, analysis, and interpretation of data; [2] drafting the paper and revising it critically; [3] approval of the submitted and final versions.

Philip Malloy: [1] substantial contributions to research design, analysis, and interpretation of data; [2] revising the paper critically; [3] approval of the submitted and final versions.

Sage Zonner: [1] substantial contributions to the acquisition, analysis, and interpretation of data; [2] drafting the paper; [3] approval of the submitted and final versions.

Cameron Gerhold: [1] substantial contributions to the acquisition, analysis, and interpretation of data; [2] drafting the paper; [3] approval of the submitted and final versions.

Eric Hu: [1] substantial contributions to the acquisition, analysis, and interpretation of data; [2] drafting the paper; [3] approval of the submitted and final versions.

Jesus Cervantes: [1] substantial contributions to the acquisition, analysis, and interpretation of data; [2] drafting the paper; [3] approval of the submitted and final versions.

Jorge Chahla: [1] substantial contributions to the interpretation of data; [2] revising the paper critically; [3] approval of the submitted and final versions.

Shane J. Nho: [1] substantial contributions to the interpretation of data; [2] revising the paper critically; [3] approval of the submitted and final versions.

Steven P. Mell: [1] substantial contributions to research design, the acquisition, analysis, and interpretation of data; [2] drafting the paper and revising it critically; [3] approval of the submitted and final versions.

All authors have read and approved the final submission of this manuscript.

AI Disclosure Statement

OpenAI's ChatGPT 5.5 was used during manuscript revision to assist with language editing and refinement. All AI‐assisted text was reviewed, edited, and verified by the authors, who remain fully responsible for the manuscript's accuracy, originality, citations, analyses, interpretation, and final content.

Data Availability Statement

The data that support the findings of this study are available on request from the corresponding author. The data are not publicly available due to privacy or ethical restrictions.