Objective:

To evaluate the accuracy of an AI model in estimating gestational age from blind sweep ultrasonography in new clinical settings.

Approach:

• Study Design: A prospective multicenter diagnostic study was conducted with 2,043 patients enrolled at urban medical centers in Chicago and Nairobi.
• Participants: 385 patients with pregnancies between 16 and 36 weeks were included in the primary analysis.
• Methods: Novice operators performed blind sweep ultrasonography using Clarius probes, and AI estimates were compared to standard clinical ultrasonography by expert sonographers.
• Primary Outcome: The mean absolute error (MAE) of gestational age estimation compared to the clinical standard.

Key Findings:

• The AI model achieved an MAE of 4.2 days, compared to 4.5 days for the clinical standard.
• Performance was consistent across both sites, with MAEs of 4.1 days in Chicago and 4.3 days in Nairobi.
• No systematic bias was observed in gestational age estimation by the AI model.
• Performance remained stable across different reference dating windows.
• The AI model showed improved accuracy for fetuses above the 90th weight percentile.

Interpretation:

The AI model demonstrated noninferior performance to the clinical standard in estimating gestational age across different clinical settings.

Limitations:

• The model was adapted using data only from Chicago prior to external validation.
• Generalizability was evaluated using a single new ultrasound manufacturer.
• Differences in operator training between sites may have influenced results.
• Subgroup analyses were based on relatively small sample sizes.

Sources:

• JAMA Network Open