Related Experiment Video
Updated: Jun 11, 2026
Automated Joint Space Detection Improves Bone Segmentation Accuracy
Published on: November 28, 2025
Compressed sensing magnetization-prepared 2 rapid gradient-echo for accelerated hip cartilage T1 mapping and deep
Julien Hirt1, Jose Roshardt2, Malin Kristin Meier3
1Department of Orthopaedic Surgery and Traumatology, Fribourg Cantonal Hospital, University of Fribourg, Fribourg, Switzerland; Department of Orthopaedic Surgery and Traumatology, Inselspital Bern, University of Bern, Bern, Switzerland.
Background:
Comprehensive assessment of hip cartilage morphology and composition is essential for surgical decision-making in femoroacetabular impingement and hip dysplasia. Magnetization-prepared 2 rapid gradient-echo (MP2RAGE) enables T1 mapping and segmentation but requires long acquisition times, limiting clinical adoption. To validate a compressed sensing (CS) MP2RAGE protocol against the standard sequence for hip cartilage T1 mapping and automated 3D segmentation of hip cartilage and labrum.
Methods:
This retrospective study included 18 patients who underwent direct MR arthrography with both standard MP2RAGE (7 min 32 s) and CS-MP2RAGE (5 min). Automated segmentation of cartilage and labrum was performed using a 3D U-Net deep learning model to extract global and regional (anterior, superior, posterior) hip cartilage T1 values and 3D parameters of hip cartilage and labrum. Morphological damage was scored semi-quantitatively on clinical proton-density weighted MRI. Agreement was assessed using paired t-tests, Bland-Altman analysis, Dice similarity coefficients, and Spearman correlations.
Results:
The mean difference in overall T1 values was 24.5 ms (4.2%, p < 0.001) with excellent correlation (rs = 0.983). Both sequences showed strong inverse correlations with morphological damage scores (CS: rs = -0.78; standard: rs = -0.76). Relative differences for cartilage volume, thickness, area, and surface area ranged from -6.9% to -0.6%. Labral volume, area, and joint surface contribution ranged from -2.6% to 0.4%. All morphological parameters showed excellent correlation (rs > 0.96).
Conclusions:
Compressed sensing MP2RAGE enables 5-min acquisition with strong agreement for cartilage T1 and automated morphological assessment compared with conventional MP2RAGE. Further prospective validation is needed before broader clinical implementation.
