Prospective Deployment of Deep Learning Reconstruction Facilitates Highly Accelerated Upper Abdominal MRI

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Conventional T1 volumetric interpolated breath-hold examination (VIBE) with SPectral Attenuated Inversion Recovery (SPAIR) fat saturation is a standard for upper abdominal MRI.
• Accelerated MRI sequences aim to reduce scan times and improve patient comfort without compromising diagnostic quality.

Purpose of the Study:

• To compare the image quality and diagnostic confidence of a deep learning (DL)-reconstructed accelerated VIBE-SPAIR sequence with a conventional VIBE-SPAIR sequence.
• To evaluate the impact of a 50% reduction in breath-hold duration on MRI outcomes.

Main Methods:

• A prospective study involving 45 patients undergoing upper abdominal MRI on a 1.5 T scanner.
• Comparison of conventional VIBE-SPAIR (16s acquisition) with DL-reconstructed VIBE-SPAIR (8s acquisition) using quantitative and qualitative image analysis by four readers.
• Assessment of fat suppression homogeneity, SNR, edge sharpness, artifacts, lesion detectability, and diagnostic confidence.

Main Results:

• The DL-reconstructed VIBE-SPAIR sequence demonstrated significantly superior SNR, edge sharpness, and reduced artifacts compared to the conventional sequence (P < 0.001).
• Interreader agreement for image quality parameters was substantial to almost perfect.
• No significant differences were observed in fat suppression homogeneity, lesion detectability, or diagnostic confidence between the two sequences (P > 0.05).

Conclusions:

• Deep learning reconstruction of accelerated VIBE-SPAIR MRI offers a significant improvement in image quality and artifact reduction.
• This accelerated technique enables a 50% reduction in breath-hold time without compromising diagnostic performance for upper abdominal imaging.