Hierarchical Attentive Network for Gestational Age Estimation in Low-Resource Settings

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Maternal-Fetal Medicine

Background:

• Fetal growth restriction (FGR) poses significant risks, particularly in low- and middle-income countries (LMICs).
• Limited access to affordable diagnostic tools hinders timely assessment of fetal development and FGR in LMICs.
• Existing healthcare barriers exacerbate maternal and fetal health issues in these regions.

Purpose of the Study:

• To introduce an end-to-end algorithm for a low-cost Doppler ultrasound device to estimate gestational age (GA) and infer FGR.
• To address the need for accessible and affordable diagnostic technologies in resource-limited settings.
• To improve the assessment of fetal development and identify potential FGR.

Main Methods:

• A hierarchical deep sequence learning model with an attention mechanism was designed.
• The model was trained on Doppler ultrasound signals from 226 pregnancies in highland Guatemala.
• Data collection was performed by lay midwives, focusing on fetal cardiac activity dynamics.

Main Results:

• The algorithm achieved state-of-the-art GA estimation performance with an average error of 0.79 months.
• The model demonstrated near-theoretical minimum error for the given data quantization.
• When applied to low birth weight fetuses, the estimated GA was lower than that from last menstruation, suggesting potential FGR.

Conclusions:

• The developed algorithm offers a promising, low-cost solution for estimating GA and detecting FGR.
• This technology can aid in identifying fetuses at risk for developmental retardation in resource-limited settings.
• Early identification of FGR can facilitate necessary referrals and interventions, improving maternal and fetal outcomes.