Deep learning creates high angular resolution diffusion data from low b-value acquisition to halve scan time for pediatric neurodevelopment imaging

In a retrospective study of 95 children, deep learning created high-b-value diffusion data from b=750, matching acquired HARDI patterns and showing consistent sex differences across tracts, potentially halving pediatric scan time.

• Retrospective study of 95 children aged 2-10 years (49 females); model trained and validated on 12 subjects each, tested on 71 subjects for tract-based analysis.
• Prediction-integrated datasets showed consistent sex differences in tract volumes compared to source-only data, except in the pyramidal tract.
• Limited training sample (n=12) and lack of external validation may affect generalizability.

RadPigeon summaries are original and for information only. They are not clinical advice.