A mobile-optimized artificial intelligence system for gestational age and fetal malpresentation assessment.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Global Health

Background:

• Fetal ultrasound is crucial for antenatal care but limited in low-to-middle-income countries due to a healthcare worker shortage.
• This study explores artificial intelligence (AI) applications for fetal ultrasound in under-resourced areas.

Purpose of the Study:

• To investigate the efficacy of AI models in predicting fetal gestational age (GA) and malpresentation using blind sweep ultrasounds.
• To assess the performance of AI models compared to standard methods and evaluate their feasibility on mobile devices.

Main Methods:

• AI models were developed using blind sweep ultrasounds collected by sonographers and novice operators in the USA and Zambia.
• AI models predicted GA and fetal malpresentation; performance was compared to standard fetal biometry and sonographer assessment.
• On-device AI model run-times were benchmarked on Android mobile phones.

Main Results:

• AI model's GA estimation accuracy was non-inferior to standard fetal biometry (error difference -1.4 ± 4.5 days).
• Non-inferiority in GA estimation was maintained even with novice operators.
• Fetal malpresentation prediction achieved a high AUC-ROC of 0.977, with similar performance across operators and devices. AI models ran in under 3 seconds on mobile phones.

Conclusions:

• AI models for GA estimation and fetal malpresentation are effective and comparable to current clinical standards.
• These AI tools can operate on-device without internet, aiding less experienced operators in resource-limited settings.
• The study demonstrates AI's potential to enhance fetal ultrasound accessibility and quality in global antenatal care.