Age-related prevalence and imaging characteristics of the superior acetabular roof notch in children, adolescents, and young adults

Institutional review board approval was obtained for this retrospective study. The local RIS and PACS databases of a tertiary general hospital and a children’s hospital serving more than 1,000,000 individuals were queried for hip MRI examinations with and without intravenous or intra-articular contrast-agent administration during an 11-year time period from 2010 to 2020. Patients without informed consent were excluded as well as patients younger than 4 years old and patients older than 25 years old. It was hypothesized that due to skeletal immaturity, a potential SARN may not be visible in the first few years of life. During the review of our data, the youngest patient with a clearly visible SARN on MRI was 4 years old, and this age was chosen as the lower inclusion age limit. The upper inclusion age limit was chosen based on the knowledge that the last growth plate of the clavicle closes in the mid-20 s and that ossification of the hip is complete between the ages of 20 and 25 years [7, 8]. Hip conditions comprising the articular cartilage and osseous structures of the hip joints were excluded, particularly, inflammatory arthritis (n = 1), osteoarthritis (n = 2), developmental dysplasia of the hip (n = 59), slipped capital femoral epiphysis (n = 2), and Perthes disease (n = 5). Moreover, MRI examinations with inappropriate image quality due to motion artifacts (n = 4) and MR images without patient’s informed consent (n = 5) were excluded. Out of 323 patients fulfilling these conditions, a total of 19 MRI examinations of patients with multiple MRI examinations were excluded. Finally, 304 patients were included, with a mean patient age of 17.9 ± 4.6 years and an age range of 4 to 24 years (Table 1). A total of 146 (48%) were female patients and 158 (52.0%) were male patients. MRI images of either the right or left hip were retrieved in 250 patients. MRI images of both hips were retrieved in 54 patients. Patients underwent MRI examinations between October 8, 2010, and September 30, 2020. If pelvic radiographs were available prior or after 180 days of the corresponding MRI, then the radiographs were included in the analysis of diagnostic accuracy (sensitivity and specificity). Thus, a total of 133 radiographs were included in the diagnostic accuracy analysis.

MR images of all patients were obtained with a 1.5-T or 3-T MRI scanner (Magnetom Avanto fit, Magnetom Aera, Magnetom Skyra, Magnetom Skyra fit, Magnetom Vida, Siemens Healthineers, Erlangen, Germany). Various standardized MRI protocols with or without intravenous or intra-articular contrast-agent administration were used, depending on the specific clinical scenario. In general, MRI protocols included T1-weighted sequences without fat suppression as well as fluid-sensitive sequences such as STIR images, T2-weighted, or intermediate-weighted images with or without fat suppression in multiple imaging planes. For example, the 3-T MR arthrography protocol (Magnetom Skyra fit) includes the following specific sequences: transverse turbo spin-echo (TSE) T1 weighted: repetition time ms [TR]/echo time ms [TE], 742/24; field of view [FOV], 180 mm; flip angle, 120°; coronal TSE T1-weighted: TR/TE, 550/12; FOV, 160 mm; flip angle, 135°; coronal TSE intermediate weighted with fat suppression: TR/TE, 4540/33; FOV, 160 mm; flip angle, 150°; sagittal TSE intermediate weighted with fat suppression: TR/TE, 4600/33; FOV, 160 mm; flip angle, 150°; transverse oblique 3D water-excitation true fast imaging with steady-state precession: TR/TE, 11.7/5.85; FOV, 160 mm; flip angle, 30°.

Digital radiography systems (DigitalDiagnost VM, Koninklijke Philips N.V., Best, The Netherlands) with a 35 × 43 cm flat panel detector were used to obtain pelvic radiographs in anteroposterior projection with the patient in the supine position. The detector-to-tube distance was 115 cm. Grid use, tube voltage, and tube current were individually adjusted to patients age, height, and weight.

The imaging characteristics of the SARN on radiographs have been described by Teichert and Johnstone et al. [3, 5]. As mentioned above, the MR imaging characteristics of the SARN have not been reported in the literature so far. On radiographs, the SARN has been reported as a discontinuity, a linear structure slightly converging in the superomedial direction, and a notch of the superomedial quadrant of the acetabulum [3, 5]. Imaging examples of superior acetabular roof notches on radiographs are shown in Fig. 1. According to the radiographic imaging characteristics, the present study identified the SARN on MRI as a discontinuous, linear structure slightly converging in the superomedial direction and notch of the superomedial quadrant of the acetabulum. Fluid-like findings within the SARN on MRI were classified as SARN type-1, whereas SARNs with fat-like findings within the SARN on MRI were classified as SARN type-2. Imaging examples of superior acetabular roof notches on MRI are shown in Figs. 2 and 3.

In the present study, MRI was considered the reference standard for diagnosing SARN on radiographs to determine the sensitivity and specificity for detecting SARN findings on radiographs.

Two observers (DV, fourth year radiology resident; TF, fellowship-trained musculoskeletal radiologist with 8 years of radiology experience and 2 dedicated years in musculoskeletal radiology) independently evaluated all images. Disagreements between the two observers were resolved by a third observer (TJD, fellowship-trained musculoskeletal radiologist with 10 dedicated years of experience in musculoskeletal radiology).

Cohen’s kappa and 95% bootstrapped percentile confidence intervals (CI) with 1000 replicates were analyzed as measures of reliability. Logistic regression models were run to assess the influence of age on the presence of SARN type-1 and SARN type-2. Influential values were checked by plotting Cook’s distance. Assumptions of linearity, normality, and homoskedasticity were checked with residual plots. Sensitivity and specificity with corresponding exact 95% confidence intervals for binomial probabilities were calculated for MRI.

A statistician (NG) performed all statistical analyses using the R programming language (R Core Team, 2020, version 4.0.2) [9]. The packages “tableone,” “irr,” “boot,” and “Hmisc” were used to compute descriptive statistics, Cohen’s kappa, confidence intervals for Cohen’s kappa, as well as for sensitivity and specificity [10‐13].