Enhancing resolution and image quality in musculoskeletal MRI using deep learning reconstruction

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Musculoskeletal (MSK) magnetic resonance imaging (MRI) is crucial for diagnosing injuries and disorders.
• Deep learning-based noise reduction offers a way to enhance image quality, balancing acquisition time, spatial resolution, and signal-to-noise ratio (SNR).

Purpose of the Study:

• To implement deep learning reconstruction (DLR) in a 1.5-T MSK MRI protocol.
• To evaluate if DLR can improve image quality, specifically spatial resolution, without compromising SNR or contrast-to-noise ratio (CNR).

Main Methods:

• Retrospective analysis of 39 MRI examinations on a 1.5-T scanner.
• Comparison of standard-resolution (SR) sequences with higher-resolution DLR (HR-DLR) sequences for knee, shoulder, ankle, and hip joints.
• Blind evaluation of structure visibility using a 5-point Likert scale by radiologists, with SNR and CNR measurements by a fourth reader.

Main Results:

• HR-DLR sequences exhibited smaller pixel size and shorter acquisition times compared to SR.
• Radiologist agreement was high for both SR and HR-DLR sequences, with higher agreement for HR-DLR.
• Likert scores for structure visibility were significantly higher or similar for HR-DLR sequences (p < 0.001).
• Apparent SNR and CNR were comparable between HR-DLR and SR sequences.

Conclusions:

• Deep learning reconstruction (DLR) effectively enhances resolution in 1.5-T MSK MRI.
• DLR improves MSK structure visualization while maintaining essential image quality metrics like SNR and CNR.
• The integration of DLR increases the efficiency of MSK MRI examinations without compromising diagnostic quality.