Accelerated Synthetic MRI with Deep Learning-Based Reconstruction for Pediatric Neuroimaging

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Pediatric Radiology

Background:

• Synthetic MRI is time-efficient but scan duration is challenging for pediatric patients.
• Accelerated techniques are needed to improve feasibility in children.
• Deep learning offers potential for faster image reconstruction.

Purpose of the Study:

• Evaluate clinical feasibility of accelerated synthetic MRI using deep learning in pediatric neuroimaging.
• Assess the impact of deep learning reconstruction on image quality.
• Investigate effects on quantitative values in synthetic MRI.

Main Methods:

• 47 children (2.3-14.7 years) underwent standard and accelerated 3T synthetic MRI.
• Accelerated scans utilized a deep learning reconstruction pipeline.
• Compared image quality, lesion detectability, tissue values, and brain volumetry.

Main Results:

• Deep learning reconstruction significantly improved accelerated scan image quality (P < .001).
• Image quality was comparable or superior to standard scans.
• No significant difference in lesion detectability (P > .05).
• Excellent agreement and strong linear relationships for tissue values and brain volumetry (R² > 0.9).

Conclusions:

• Deep learning-based reconstruction in synthetic MRI reduces scan time by 42% while preserving image quality and quantitative accuracy.
• Accelerated deep learning synthetic MRI can replace standard synthetic MRI for contrast-weighted and quantitative imaging.
• This method enhances the clinical utility of synthetic MRI in pediatric neuroimaging.