3D-QALAS synthetic MRI with Zero-DeepSub in children: initial experience including post-contrast imaging feasibility.

Area of Science:

• Medical Imaging
• Pediatric Radiology
• Artificial Intelligence in Medicine

Background:

• Pediatric MRI exams are lengthy and prone to motion artifacts due to multiple required sequences.
• Synthetic MRI techniques aim to generate multiple image contrasts from a single acquisition.
• 3D-quantification using interleaved Look-Locker acquisition sequence with T2 preparation pulse (3D-QALAS) combined with deep learning (Zero-DeepSub) enables high-resolution imaging.

Purpose of the Study:

• To evaluate the diagnostic performance of synthetic brain MRI images generated by 3D-QALAS with Zero-DeepSub reconstruction.
• To compare these synthetic images against conventional MRI sequences in pediatric patients.
• To assess image quality and the detection of specific imaging findings.

Main Methods:

• Prospective study of 26 pediatric patients undergoing brain MRI.
• Generation of synthetic T1-weighted, T2-weighted, and FLAIR images from 3D-QALAS with Zero-DeepSub.
• Independent assessment of seven predefined imaging findings by two neuroradiologists, with adjudication for discrepancies.
• Semiquantitative image quality evaluation using a 5-point Likert scale.

Main Results:

• Synthetic images demonstrated high sensitivity and specificity for detecting lesions, encephalomalacia, and collections with excellent interobserver agreement.
• Moderate specificity was noted for gliosis detection by one reader.
• No significant difference in image quality was found between synthetic and conventional FLAIR, T1-weighted, or post-contrast images (P>0.1).
• Conventional T2-weighted imaging was found to be significantly superior (P<0.001).

Conclusions:

• 3D-QALAS with Zero-DeepSub reconstruction successfully synthesizes high-resolution, clinically interpretable pediatric brain images, including post-contrast sequences.
• Synthetic contrasts were comparable to conventional images, except for T2-weighted imaging, where conventional sequences remained superior.
• This technique has the potential to shorten pediatric neuroimaging scan times, pending further optimization and validation.