Oral Concurrent Session 3 - Fetus and Fetal Intervention
Oral Concurrent Sessions
Anna Madden-Rusnak, PhD (she/her/hers)
Research Scientist
Department of Women’s Health, Dell Medical School at the University of Texas at Austin
Austin, Texas, United States
Kenneth J. Moise, Jr., MD (he/him/his)
Co-Director, Comprehensive Fetal Care Center; Prof, Women’s Health
Department of Women’s Health, Dell Medical School at the University of Texas at Austin
Austin, Texas, United States
Kelly P. Gaither, PhD
Senior Research Scientist; Associate Prof, Women's Health
Texas Advanced Computing Center
Austin, Texas, United States
Kathy J. Lowry, MSN, RN
Senior RN Clinical Research Coordinator
Department of Women’s Health, Dell Medical School at the University of Texas at Austin
Austin, Texas, United States
Emily Hutson, MSN, RN
Research Associate/Clinical Outcomes & Quality Data Specialist
Department of Women’s Health, Dell Medical School at the University of Texas at Austin
Austin, Texas, United States
Danielle Bruns, BS, RDMS
Fetal Sonographer
Department of Women’s Health, Dell Medical School at the University of Texas at Austin
Austin, Texas, United States
Reinaldo Valero, MD, RDMS
Maternal Fetal Medicine Ultrasound Technician
Department of Women’s Health, Dell Medical School at the University of Texas at Austin
Austin, Texas, United States
Reducing late-term stillbirth remains challenging due to costly fetal monitoring and unreliable patient perception of fetal movement. This study explores a low-cost approach using smartphone audio recordings and Artificial Intelligence (AI) to provide objective, reliable fetal movement (FM) metrics.
Study Design:
This longitudinal cohort study enrolled pregnant patients (≥22 weeks GA) in a Southern US city for up to five 30-minute sessions each. Fetal movements (gross, breathing, hiccups) were recorded via smartphone audio from the abdomen, ultrasound, and patient perception. Pre-processed acoustic features were analyzed to assess effects of BMI and GA. PERMANOVA compared GA groups (early: 22–27w; middle: 28–32w; late: ≥33w) and BMI categories (normal < 25; overweight 25–29.9; obese ≥30). The AI model combined machine learning and deep learning, using gradient boosting to improve FM classification. Ultrasound-detected FMs (total and average per session) and AI vs patient accuracy are reported by FM type.
Results:
Thirty patients (87% White, 53% multiparous) with singleton pregnancies were enrolled. Most (70%) completed all 5 visits (range 3–5), totaling 2,070 minutes of data. GA and BMI significantly affected audio features (p < 0.001), with stronger GA effects between early and middle groups. BMI remained mostly stable across GA within patients, but acoustic features differed across BMI categories. AI models outperformed patient perception for all fetal movement types: ~5x more accurate for gross movements, ~24x for breathing, and ~3x for hiccups.
Conclusion:
AI detection of FM from smartphone audio is both feasible and more reliable than patient perceptions. Given the ubiquity and affordability of smartphones, this approach offers a scalable, low-cost solution for at-home fetal monitoring. This approach provides an objective tool to detect changes in fetal activity that may precede late stillbirth. By leveraging accessible devices, this application holds promise for reducing disparities in prenatal care access, particularly in underserved or resource-limited populations.