Clinical validation of synthetic brain MRI in children: initial experience

The purpose of this study was to determine the diagnostic accuracy of synthetic MR sequences generated through post-acquisition processing of a single sequence measuring inherent R1, R2, and PD tissue properties compared with sequences acquired conventionally as part of a routine clinical pediatric brain MR exam.

Thirty-two patients underwent routine clinical brain MRI with conventional and synthetic sequences acquired (22 abnormal). Synthetic axial T1, T2, and T2 fluid attenuation inversion recovery or proton density-weighted sequences were made to match the comparable clinical sequences. Two exams for each patient were de-identified. Four blinded reviewers reviewed eight patients and were asked to generate clinical reports on each exam (synthetic or conventional) at two different time points separated by a mean of 33 days. Exams were rated for overall and specific finding agreement (synthetic/conventional and compared to gold standard consensus review by two senior reviewers with knowledge of clinical report), quality, and diagnostic confidence.

Overall agreement between conventional and synthetic exams was 97%. Agreement with consensus readings was 84% (conventional) and 81% (synthetic), p = 0.61. There were no significant differences in sensitivity, specificity, or accuracy for specific imaging findings involving the ventricles, CSF, brain parenchyma, or vasculature between synthetic or conventional exams (p > 0.05). No significant difference in exam quality, diagnostic confidence, or noise/artifacts was noted comparing studies with synthetic or conventional sequences.

Diagnostic accuracy and quality of synthetically generated sequences are comparable to conventionally acquired sequences as part of a standard pediatric brain exam. Further confirmation in a larger study is warranted.