Text

Objectives and questions: Radiographs and CT scans are standard for diagnosing hip deformities before hip-preserving surgery but involve ionizing radiation, raising concerns about cumulative exposure, particularly in younger patients. MRI offers detailed, radiation-free imaging of bone morphology and soft tissues. Recent advances in artificial intelligence enable automatic segmentation and MRI-based 3D pelvic modeling, allowing the creation of virtual radiographic-like projections.

The objective of this study was to determine whether standard radiographic parameters—lateral center-edge angle (LCEA), retroversion index (RI), and acetabular index (AI)—can be accurately measured on MRI-based radiographic-like projections and yield results comparable to conventional anteroposterior (AP) pelvic radiographs.

Material and methods: Fifty consecutive patients (100 hips, mean age 28 years) with symptomatic hip deformities underwent comprehensive radiologic evaluation, including standard AP pelvic radiographs and direct hip MR arthrography (T1-w VIBE DIXON, 1 mm slice thickness) as part of our institutional routine protocol. The osseous pelvis was segmented using a validated deep learning method (3D nnU-Net), and radiograph-like projections were generated using an adapted cone beam projection algorithm (MATLAB). The arthrography and non-arthrography sides were analyzed separately to assess potential measurement differences. Standard radiographic parameters were measured on MRI-based radiographic-like projections and conventional radiographs using Hip2Norm software. T-tests and Pearson's correlation were conducted for statistical analysis.

Results: The LCEA showed no significant difference between conventional and MRI-based projections, with absolute mean differences of 0.18±1.15° (arthrography, p=0.71) and 0.59±1.69° (non-arthrography, p=0.29). Similarly, the AI showed no significant difference (0.60±1.52° (arthrography, p=0.20); 1.01±2.12° (non-arthrography, p=0.075)). For the RI, absolute mean differences were 0.46±4.00% (arthrography, p=0.80) and 2.04±5.33% (non-arthrography, p=0.22), indicating no significant difference. Moderate to strong correlations were found between MRI-based and conventional radiographs (LCEA: r=0.84/0.77, AI: r=0.75/0.66, RI: r=0.60/0.66 for arthrography/non-arthrography; all p<0.001).

Discussion and conclusions: Irrespective of contrast agent use during hip MRI, our approach utilizing T1-w VIBE DIXON and a deep learning-based image post-processing method enables the creation of radiographic-like projections comparable to standard AP pelvic radiographs for the evaluated parameters. This method demonstrates high potential for assessing osseous hip morphology.