Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Point-of-Care Real-Time Signal Quality for Fetal Doppler Ultrasound Using a Deep Learning Approach

Mohsen Motie-Shirazi, Reza Sameni, Peter Rohloff, Nasim Katebi, Gari D. Clifford

TL;DR

This work addresses the challenge of variable signal quality in 1D-Doppler fetal ultrasound collected with low-cost smartphone systems in rural, Indigenous communities. It develops a scalogram-based deep learning pipeline (CNN+GRU+Attention) that operates on $3.75\,\mathrm{s}$ windows to classify signal quality into five categories, achieving a micro F1 of $97.41\%$ and a macro F1 of $94.23\%$, with the Good class exceeding $F1=99\%$. Compared with a prior SVM-based approach on the same data, the deep learning model shows superior performance for Good and Poor signals, while SVM is relatively stronger for Interference. The model is designed for on-device deployment via TensorFlow Lite, enabling real-time feedback to traditional midwives and pointing to future multi-task extensions (e.g., gestational age, fetal heart rate) and iterative learning to adapt to new data and environments.

Abstract

In this study, we present a deep learning framework designed to integrate with our previously developed system that facilitates large-scale 1D fetal Doppler data collection, aiming to enhance data quality. This system, tailored for traditional Indigenous midwives in low-resource communities, leverages a cost-effective Android phone to improve the quality of recorded signals. We have shown that the Doppler data can be used to identify fetal growth restriction, hypertension, and other concerning issues during pregnancy. However, the quality of the signal is dependent on many factors, including radio frequency interference, position of the fetus, maternal body habitus, and usage of the Doppler by the birth attendants. In order to provide instant feedback to allow correction of the data at source, a signal quality metric is required that can run in real-time on the mobile phone. In this study, 191 DUS signals with durations mainly in the range between 5 to 10 minutes were evaluated for quality and classified into five categories: Good, Poor, (Radiofrequency) Interference, Talking, and Silent, at a resolution of 3.75 seconds. A deep neural network was trained on each 3.75-second segment from these recordings and validated using five-fold cross-validation. An average micro F1 = 97.4\% and macro F1 = 94.2\% were achieved, with F1 = 99.2\% for `Good' quality data. These results indicate that the algorithm, which will now be implemented in the midwives' app, should allow a significant increase in the quality of data at the time of capture.

Point-of-Care Real-Time Signal Quality for Fetal Doppler Ultrasound Using a Deep Learning Approach

TL;DR

This work addresses the challenge of variable signal quality in 1D-Doppler fetal ultrasound collected with low-cost smartphone systems in rural, Indigenous communities. It develops a scalogram-based deep learning pipeline (CNN+GRU+Attention) that operates on windows to classify signal quality into five categories, achieving a micro F1 of and a macro F1 of , with the Good class exceeding . Compared with a prior SVM-based approach on the same data, the deep learning model shows superior performance for Good and Poor signals, while SVM is relatively stronger for Interference. The model is designed for on-device deployment via TensorFlow Lite, enabling real-time feedback to traditional midwives and pointing to future multi-task extensions (e.g., gestational age, fetal heart rate) and iterative learning to adapt to new data and environments.

Abstract

In this study, we present a deep learning framework designed to integrate with our previously developed system that facilitates large-scale 1D fetal Doppler data collection, aiming to enhance data quality. This system, tailored for traditional Indigenous midwives in low-resource communities, leverages a cost-effective Android phone to improve the quality of recorded signals. We have shown that the Doppler data can be used to identify fetal growth restriction, hypertension, and other concerning issues during pregnancy. However, the quality of the signal is dependent on many factors, including radio frequency interference, position of the fetus, maternal body habitus, and usage of the Doppler by the birth attendants. In order to provide instant feedback to allow correction of the data at source, a signal quality metric is required that can run in real-time on the mobile phone. In this study, 191 DUS signals with durations mainly in the range between 5 to 10 minutes were evaluated for quality and classified into five categories: Good, Poor, (Radiofrequency) Interference, Talking, and Silent, at a resolution of 3.75 seconds. A deep neural network was trained on each 3.75-second segment from these recordings and validated using five-fold cross-validation. An average micro F1 = 97.4\% and macro F1 = 94.2\% were achieved, with F1 = 99.2\% for `Good' quality data. These results indicate that the algorithm, which will now be implemented in the midwives' app, should allow a significant increase in the quality of data at the time of capture.
Paper Structure (16 sections, 5 equations, 2 figures, 5 tables)

This paper contains 16 sections, 5 equations, 2 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Methods
  3. Data Collection
  4. Annotation of DUS Recordings
  5. Data Preparation
  6. Network Architecture
  7. Model Implementation and Evaluation
  8. Comparison with Prior Work
  9. Results
  10. Model Assessment
  11. Evaluating Deep Learning and SVM on Common Data
  12. Discussion
  13. Performance and Limitations of the CNN+GRU+Att Model
  14. Deep Learning vs Support Vector Machine in Quality Classification
  15. Implications and Future Directions
  16. ...and 1 more sections

Figures (2)

  • Figure 1: Histogram of the DUS recording duration. Note that the signal was broken into 5 minute segments (in case of file corruption) and later reconstituted into single file. This explains the periodicity in the distribution.
  • Figure 2: Architecture of the deep learning model consisting of CNN and GRU layers, followed by an attention mechanism.