Deep learning-enhanced zero echo time MRI for femoral cortical bone quantification: A technical validation against CT

This study compared quantitative measurements of femoral cortical bone between zero echo time magnetic resonance images (oZTEo) and computed tomographic (CT) images as proof-of-concept for algorithm performance in subjects referred to our clinic for femoroacetabular impingement (FAI), a morphological condition of the hip joint associated with early onset osteoarthritis. A total of 30 patients underwent concomitant MRI and CT imaging where standard oZTEo imaging was performed with deep learning (DL) reconstruction to apply denoising and chemical shift correction (oZTEo-CSC) with and without automated background removal (oZTEo-CSC-BR). The CT images were registered and resampled into the corresponding MR image space. A normalized signal intensity profile analysis was performed across the articular surface of the femoral head to evaluate: intra-articular femoral cortex edge location, cortical edge sharpness, maximum cortical signal intensity, and cortical thickness. A one-way repeated measures of variance (ANOVA) was used to compare outcomes across image sets (CT, oZTEo, oZTEo-CSC, and oZTEo-CSC-BR) and intra-class correlation coefficients (ICCs) were assessed for inter-examiner repeatability. The intra-articular femoral cortex edge and the cortical edge sharpness did not vary across image datasets for either Examiner. Maximum cortical signal intensity differed across reconstructions for Examiner 1 (p < 0.0001), with oZTEo differing from CT, oZTEo-CSC, and oZTEo-CSC-BR methods, while the oZTEo-CSC and oZTEo-CSC-BR reconstructions were not different from CT. The oZTEo-CSC and oZTEo-CSC-BR images provide a greater CT-like assessment of femoral cortical bone than oZTEo images alone and may be considered as a radiation-free alternative for bone imaging. Inter-examiner repeatability was higher for the deep learning reconstructions (ICC ≈ 0.68) compared with oZTEo and CT, indicating improved measurement consistency across readers. oZTEo with DL post-processing improves consistency and CT agreement; differences are sub-voxel and clinically negligible overall.