Deep Learning Reconstruction Enables Diagnostic-Quality 0.4T Knee and Spine MRI in One-Third of the Time

Clinical Snapshot

Key Findings

• P Value: Not reported for primary comparisons

• Effect Size: Kendall's Tau correlation values: 0.64 (image quality), 0.64 (sharpness), 0.54 (anatomical conspicuity), 0.55 (artifacts)

• Primary Outcome: Radiologist assessment scores for image quality, sharpness, anatomical conspicuity, and artifacts across acceleration factors 1.5-4

• Nnt Or Sensitivity: Inter-observer agreement: Cohen's kappa 0.227-0.299 (poor to fair agreement)

• Confidence Interval: Not reported for primary outcomes

Clinical Application

Requires implementation of deep learning reconstruction software on 0.4T MRI systems, which may have cost and technical training implications Relevant for rural and remote Australian healthcare settings where MRI access is limited. Could support PBS item 63560 (knee MRI) efficiency, though TGA approval would be required for clinical implementation Patients requiring knee or spine MRI where faster acquisition is beneficial, particularly in high-throughput settings or claustrophobic patients

Abstract

There has been a growing interest in low-field MRI due to its lower costs, enabling an increase in accessibility of MRI worldwide. Long scan times are currently needed to obtain high-quality images. The aim of our study was to accelerate low-field 0.4T musculoskeletal MRI using deep learning and evaluate maximum acceleration via assessment by two expert radiologists. A cascades of independently recurrent inference machines (CIRIM) model was trained using knee and spine 2D multislice scans, retrospectively undersampled in k-space. The heterogeneous dataset included varied contrasts, orientations, and matrix sizes. To determine the maximum possible acceleration factor, the undersampling pattern and loss function of the CIRIM were optimized using the knee dataset. Models were trained for acceleration factors 1.5-4. The different acceleration factors were evaluated quantitatively and clinically, with radiologist scores compared using Kendall's Tau and Cohen's Kappa. To assess the model's generalizability, two prospectively undersampled scans were evaluated, and the knee-trained models were tested on spine data. CIRIM accurately reconstructed all data, with slight metric degradation at higher accelerations. Combining L1, structural similarity index measure (SSIM), and perceptual loss improved sharpness, while undersampling pattern differences were minimal. Clinical scores from both radiologists declined with acceleration, reflected in Kendall's Tau values of 0.64, 0.64, 0.54, and 0.55 for image quality, sharpness, anatomical conspicuity, and artifacts. Inter-observer agreement was low (Cohen's kappa: 0.227-0.299). The model generalized well, reconstructing spine data with a knee-trained model at comparable quality and successfully handling prospectively undersampled scans. CIRIM consistently delivered high-quality reconstructions, demonstrating excellent robustness to variations in contrast, orientation, matrix size, and anatomy. Its ability to generalize across anatomies and handle prospectively undersampled scans highlights its practical utility. The differences in reading for higher accelerations between radiologists suggest that the optimal acceleration factor is context-dependent and should be tailored to the specific diagnostic task.

References

1. 1.van den Berg, D. M., Varriale, R., Ferrando, F., Traverso, P., Balbi, L., Belgiorno, G., Beekman, K., Kox, L., Strijkers, G. J., & Caan, M. W. A. (2026). Deep Learning Reconstruction Enables Diagnostic-Quality 0.4T Knee and Spine MRI in One-Third of the Time. NMR in Biomedicine. https://doi.org/10.2139/ssrn.4801289