Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Development and Evaluation of an AI-Driven Telemedicine System for Prenatal Healthcare

Juan Barrientos, Michaelle Pérez, Douglas González, Favio Reyna, Julio Fajardo, Andrea Lara

TL;DR

The paper tackles the lack of prenatal ultrasound access in rural LMICs by introducing NatalIA, a human‑in‑the‑loop AI system that uses blind sweep ultrasound to help midwives capture diagnostically relevant fetal planes with remote specialist review. It integrates a web platform for asynchronous interpretation with DL preselection of standard planes from nonexpert sweeps, leveraging transfer learning on architectures like ResNet‑50 to achieve high accuracy ($92.87\%$) on phantom data. Field usability tests show the approach yields good usability and low NASA‑TLX cognitive workload, supporting potential deployment in underserved regions. Overall, the work demonstrates the feasibility and value of tele‑ultrasound with human oversight to expand equitable prenatal imaging access and support community‑based maternal care.

Abstract

Access to obstetric ultrasound is often limited in low-resource settings, particularly in rural areas of low- and middle-income countries. This work proposes a human-in-the-loop artificial intelligence (AI) system designed to assist midwives in acquiring diagnostically relevant fetal images using blind sweep protocols. The system incorporates a classification model along with a web-based platform for asynchronous specialist reviews. By identifying key frames in blind sweep studies, the AI system allows specialists to concentrate on interpretation rather than having to review entire videos. To evaluate its performance, blind sweep videos captured by a small group of soft-trained midwives using a low-cost Point-of-Care Ultrasound (POCUS) device were analyzed. The system demonstrated promising results in identifying standard fetal planes from sweeps made by non-experts. A field evaluation indicated good usability and a low cognitive workload, suggesting that it has the potential to expand access to prenatal imaging in underserved regions.

Development and Evaluation of an AI-Driven Telemedicine System for Prenatal Healthcare

TL;DR

The paper tackles the lack of prenatal ultrasound access in rural LMICs by introducing NatalIA, a human‑in‑the‑loop AI system that uses blind sweep ultrasound to help midwives capture diagnostically relevant fetal planes with remote specialist review. It integrates a web platform for asynchronous interpretation with DL preselection of standard planes from nonexpert sweeps, leveraging transfer learning on architectures like ResNet‑50 to achieve high accuracy () on phantom data. Field usability tests show the approach yields good usability and low NASA‑TLX cognitive workload, supporting potential deployment in underserved regions. Overall, the work demonstrates the feasibility and value of tele‑ultrasound with human oversight to expand equitable prenatal imaging access and support community‑based maternal care.

Abstract

Access to obstetric ultrasound is often limited in low-resource settings, particularly in rural areas of low- and middle-income countries. This work proposes a human-in-the-loop artificial intelligence (AI) system designed to assist midwives in acquiring diagnostically relevant fetal images using blind sweep protocols. The system incorporates a classification model along with a web-based platform for asynchronous specialist reviews. By identifying key frames in blind sweep studies, the AI system allows specialists to concentrate on interpretation rather than having to review entire videos. To evaluate its performance, blind sweep videos captured by a small group of soft-trained midwives using a low-cost Point-of-Care Ultrasound (POCUS) device were analyzed. The system demonstrated promising results in identifying standard fetal planes from sweeps made by non-experts. A field evaluation indicated good usability and a low cognitive workload, suggesting that it has the potential to expand access to prenatal imaging in underserved regions.

Paper Structure

This paper contains 15 sections, 5 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. System Architecture
  4. Dataset
  5. DL Models
  6. Experiments and Results
  7. DL Model Evaluation
  8. Clinical Validation of System’s Predictions by Specialists
  9. Field Evaluation Study
  10. Blind Sweep Protocol Usability
  11. System Usability
  12. Cognitive Workload Evaluation
  13. Discussion and Conclusion
  14. Acknowledgments.
  15. Disclosure of Interests.

Figures (5)

  • Figure 1: Midwives from rural areas in Guatemala performing blind sweep ultrasound scans using a handheld POCUS device.
  • Figure 2: System architecture composed of (a) backend and (b) frontend components.
  • Figure 3: Blind sweep acquisition protocol with vertical, horizontal, and two diagonal trajectories.
  • Figure 4: Confusion matrices from the transfer learning experiment for the three architectures: (a) SonoNet, (b) ResNet-18, and (c) ResNet-50. Labels indicate fetal planes: BPD (biparietal diameter), AC (abdominal circumference), HS (heart standard view), SS (spine standard view), FL (femur length), and No plane (non-diagnostic frame).
  • Figure 5: Average scores reported by participants across the six NASA-TLX workload.