Fetal brain age estimation and anomaly detection using attention-based deep ensembles with uncertainty
Wen Shi1, Guohui Yan2, Yamin Li3
1Key Laboratory for Biomedical Engineering of Ministry of Education, Department of Biomedical Engineering, College of Biomedical Engineering & Instrument Science, Zhejiang University, Hangzhou, 310027, China.
Neuroimage
|September 5, 2020
Summary
Researchers developed an AI model to predict fetal brain age using MRI scans, achieving high accuracy. This tool also detects fetal brain abnormalities, offering potential for improved prenatal diagnosis.
Area of Science:
- Neuroimaging
- Artificial Intelligence
- Medical Diagnostics
Background:
- MRI-based brain age prediction is established for postnatal development.
- Fetal brain age prediction using MRI is an underexplored area with diagnostic potential.
- Limited tools exist for assessing fetal brain development and abnormalities.
Purpose of the Study:
- To develop a novel method for accurate fetal brain age prediction using MRI.
- To create AI-driven markers for detecting fetal brain anomalies.
- To explore the clinical utility of AI in prenatal examination.
Main Methods:
- An attention-based deep residual network was trained on 659 T2-weighted fetal brain MRI scans.
- An ensemble method was used to quantify predictive uncertainty and estimation confidence.
- Attention maps were generated to visualize contributing regional features.
Main Results:
- The model achieved a mean absolute error of 0.767 weeks and R² of 0.920 for fetal brain age prediction.
- Novel uncertainty-based markers showed diagnostic power for small head circumference (AUC 0.90), malformations (AUC 0.90), and ventriculomegaly (AUC 0.67).
- Attention maps highlighted gestational stage-specific regional features important for age estimation.
Conclusions:
- Attention-based deep ensembles offer superior performance in fetal brain age estimation.
- The developed AI approach shows promise for detecting fetal brain anomalies.
- This technology has the potential for clinical translation in prenatal diagnosis.
