AI integration with point-of-care ultrasonography tied to reliable gestational age assessment in low resource settings: JAMA

Accurate assessment of gestation is paramount in the provision of appropriate care in pregnancy and for guiding interventions and monitoring fetal development. Traditionally, this mandates high-specification ultrasonography that most likely may not exist in a low-resource setting. In this study, an AI-enabled ultrasonography tool will be created for use by novice users who have no prior training in sonography.

This was a prospective diagnostic accuracy study that involved 400 women with viable, singleton, nonanomalous, first-trimester pregnancies from Lusaka, Zambia, and Chapel Hill, North Carolina. GA was determined by experienced sonographers with transvaginal crown-rump length measurements. At random follow-up visits spanning the window of the primary evaluation from 14 to 27 weeks of gestation, novice users obtained blinded sweeps of the maternal abdomen with an AI-enabled device. Comparative fetal biometry was performed on a high-specification machine by experienced sonographers.

The primary outcome was the mean absolute error of the AI-enabled device compared with the study standard. The accuracy of the device was considered equivalent if the difference in MAE fell within a prespecified margin of ±2 days. The percentage of assessments within 7 days of ground truth GA was measured secondarily.

Results

• In the primary evaluation window, the AI-enabled device demonstrated equivalence to the study standard with an MAE of 3.2 days (SE 0.1) versus 3.0 days (SE 0.1) for the standard method (difference, 0.2 days [95% CI, −0.1 to 0.5]).

• The percentage of assessments within 7 days of the ground truth GA was 90.7% for the AI device and 92.5% for the study standard.

• The AI tool's performance was consistent across various subgroups, including different geographical locations and body mass index categories.

The study epitomizes how AI technology can bridge this gap in prenatal care within low-resource settings. With the aid of an AI-enabled device, this brings novice users closer to the same level of GA estimation accuracy attained by credentialed sonographers using high-specification equipment. This is most relevant where access to advanced medical technology is limited and thus serves a universal access aim of the World Health Organization for GA estimation during pregnancy.

Novice operators who had never received ultrasonography training estimated GA for dates between 14 and 27 weeks of gestation as accurately with this low-cost, AI-enabled device as credentialed sonographers performing standard biometry on high-specification machines. These findings have immediate implications for obstetrical care in low-resource settings, and on a global scale, they have advanced global health initiatives and improved pregnancy outcomes.

Reference:

Stringer, J. S. A., Pokaprakarn, T., Prieto, J. C., Vwalika, B., Chari, S. V., Sindano, N., Freeman, B. L., Sikapande, B., Davis, N. M., Sebastião, Y. V., Mandona, N. M., Stringer, E. M., Benabdelkader, C., Mungole, M., Kapilya, F. M., Almnini, N., Diaz, A. N., Fecteau, B. A., Kosorok, M. R., … Kasaro, M. P. (2024). Diagnostic accuracy of an integrated AI tool to estimate gestational age from blind ultrasound sweeps. JAMA: The Journal of the American Medical Association. https://doi.org/10.1001/jama.2024.10770