Deep learning improves image quality in motion-robust and sedation-free pediatric brain MRI.

Area of Science:

• Radiology
• Medical Imaging
• Artificial Intelligence in Medicine

Background:

• Pediatric MRI quality is often compromised by motion and limited patient compliance, frequently necessitating sedation.
• Single-shot MRI sequences offer speed but typically yield lower image quality.
• Deep learning (DL) presents a potential solution for enhancing these rapid imaging techniques.

Purpose of the Study:

• To evaluate the diagnostic performance of a DL framework combining compressed sensing (CS) and convolutional neural networks (CNNs) for T2-weighted single-shot MRI (T2-SSH_DL).
• To compare T2-SSH_DL against conventional CS-based reconstruction (T2-SSH_conv) and standard high-resolution T2-weighted sequences.
• To assess the potential of DL to improve image quality and reduce sedation requirements in pediatric brain MRI.

Main Methods:

• Prospective single-center study of 62 pediatric patients (mean age 7.4 years).
• T2-weighted single-shot brain MRI acquired from both sedated and awake children.
• Raw data reconstructed using a DL pipeline and conventional CS; quantitative metrics (aCNR, aSNR, ERD) and qualitative radiologist assessments were performed.

Main Results:

• T2-SSH_DL demonstrated significantly higher apparent contrast-to-noise ratio (aCNR) and apparent signal-to-noise ratio (aSNR) compared to T2-SSH_conv.
• Improved image sharpness (edge rise distance) and reduced motion artifacts were observed with T2-SSH_DL.
• Qualitative assessments revealed superior overall image quality, lesion conspicuity, and sharpness for the DL-enhanced sequences.

Conclusions:

• DL-based reconstruction substantially enhances diagnostic quality of T2-weighted single-shot brain MRI in pediatric patients.
• This ultrafast, motion-robust imaging approach achieves clinically usable quality, potentially decreasing the need for sedation.
• The DL framework offers a promising method to improve patient safety and efficiency in pediatric neuroimaging.