What is the Impact of Applying Deep Learning Reconstruction to Zero Echo Time MRI on the Evaluation of Osseous Lesions in Axial Spondyloarthritis?

Commentary on Yi J et al. Deep learning reconstruction of zero echo time magnetic resonance imaging: diagnostic performance in axial spondyloarthritis. Eur Radiol 2026;36:296-307 (https://doi.org/10.1007/s00330-025-11843-3).

SUMMARY

This study evaluated the diagnostic performance of deep learning reconstruction (DLR) of zero-echo-time (ZTE) MRI for structural lesions in patients with axial spondyloarthritis (SpA), compared with T1-weighted imaging (T1WI) and ZTE MRI without DLR, using CT as the reference standard (1). Fifty-two sacroiliac (SI) joint MRIs (104 quadrants) from 26 patients were assessed by three readers for image quality and structural conspicuity, with SI joints scored for structural lesions on T1WI, ZTE, and ZTE DLR at varying intensities (50%, 75%, and 100%). ZTE DLR 100% yielded the highest image quality scores for two of the three readers (R1 and R2) and superior structural conspicuity across all readers. Regarding bone sclerosis, ZTE DLR 50% (in R1) or 75% (in R2 and R3) showed the best diagnostic performance, with the highest interreader agreement in R1 and the other two readers. Regarding bone erosion, ZTE DLR 100% showed the best diagnostic performance across all readers, with the highest interreader agreement. ZTE MRI with DLR outperformed T1WI, and ZTE MRI without DLR in diagnosing bone sclerosis and bone erosion at the SI joints involved by SpA.

COMMENT

ZTE MRI provides CT-like images; however, its relatively low signal-to-noise ratio makes it challenging to detect subtle structural lesions in the SI joints of patients with SpA (2). This study tested the hypothesis that recently developed DLR could improve the quality of ZTE MRI, potentially enabling its inclusion in standard MRI protocols as an alternative to CT. The results indicate that a higher percentage of DLR significantly improves structural conspicuity. Of particular interest is the differential diagnostic performance for specific osseous lesions, namely bone sclerosis and bone erosion. Given this difference, it may be beneficial to optimize the DLR denoising level based on the type of targeted osseous lesion. Furthermore, these results suggest that varying denoising levels may be necessary when applying ZTE DLR to different anatomic regions in the future.