A framework for gestational development tracking using 1D-Doppler ultrasound signals in rural Guatemala Open Access

Valderrama Cuadros, Camilo Ernesto (Spring 2020)

Permanent URL: https://etd.library.emory.edu/concern/etds/wm117q12q?locale=en
Published

Abstract

Guatemala shoulders the burden of one of the highest perinatal mortality rates in Latin America, particularly among rural indigenous (Mayan) communities. This thesis aims to develop a fetal monitoring system for a highland community in Chimatenalgo, Guatemala, by designing methods to analyze one-dimension Doppler ultrasound signals (1D-DUS) recorded with a low-cost mobile health system piloted in the same community. 

   

  To that end, signal processing and machine learning techniques were used to address issues found in the pilot and initial randomized control trial of the mobile health system. In particular, four related pieces of research were undertaken. First, a signal quality method was developed to asses the utility of the 1D-DUS, achieving an F1-score higher than 90\% to classify the signals into five distinct types of error. Then, an autocorrelation-based method to estimate fetal heart rate (FHR) from 1D-DUS was developed using a dataset simultaneously recorded with a fetal electrocardiogram. The method was shown to be generalizable, accurately estimating FHR for two independent datasets, including one collected in the Guatemalan highland community that is the focus of this study. Third, estimation of birth weight from observed postnatal weight was performed as a proxy to identify small-for-gestational-age births, achieving similar results to those provided by the Guatemalan government for the region of study. Fourth, fetal heart rate variability indices were combined with maternal blood pressure readings to estimate gestational age using a supervised support vector regression approach. The estimations resulted in a mean absolute estimation error of 0.8 months, which is comparable to previous works developed in industrialized environments while requiring only an inexpensive transducer and a self-inflating blood pressure device. 

   

  This thesis provides low-cost approaches for identifying high-quality 1D-DUS signals, estimating FHR, and in turn, estimating gestational age in order to identify potential cases of Low Birth Weight, Small for Gestational Age, or Intrauterine Growth Restriction. The work empowers traditional birth attendants with a decision support system to identify patients with possible pregnancy-related abnormalities requiring professional medical assistance. 

Table of Contents

1 Introduction 1

1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Aim of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3

1.3 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5

1.4 List of publication . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Background 8

2.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8

2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9

2.3 Common pregnancy complications . . . . . . . . . . . . . . . . . . . . 9

2.3.1 Intrauterine growth restriction . . . . . . . . . . . . . . . . . . 9

2.4 Traditional physical exams . . . . . . . . . . . . . . . . . . . . . . . . 11

2.4.1 Fundal height . . . . . . . . . . . . . . . . . . . . . . . . . . . 11

2.4.2 Fetal movement counting . . . . . . . . . . . . . . . . . . . . . 11

2.4.3 Maternal blood pressure monitoring . . . . . . . . . . . . . . . 11

2.5 Biochemical tests . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.5.1 Estriol assays . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2.5.2 Human placental lactogen . . . . . . . . . . . . . . . . . . . . 13

2.6 Fetal cardiac assessment . . . . . . . . . . . . . . . . . . . . . . . . . 13

2.6.1 Fetal cardiac circulation . . . . . . . . . . . . . . . . . . . . . 13

2.6.2 Control of fetal heart rate . . . . . . . . . . . . . . . . . . . . 14

2.6.3 Fetal heart monitoring techniques . . . . . . . . . . . . . . . . 15

2.7 Imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2.7.1 Ultrasound imaging . . . . . . . . . . . . . . . . . . . . . . . . 23

2.7.2 Fetal echocardiography . . . . . . . . . . . . . . . . . . . . . . 27

2.7.3 Fetal Magnetic Resonance Imaging . . . . . . . . . . . . . . . 30

2.8 Summary of fetal heart monitoring techniques . . . . . . . . . . . . . 32

2.9 Perinatal mortality and fetal monitoring in low-and middle income

countries and resource-constrained region . . . . . . . . . . . . . . . . 33

2.9.1 Mobile technology for perinatal care . . . . . . . . . . . . . . . 37

2.10 Discussion and Conclusion . . . . . . . . . . . . . . . . . . . . . . . . 40

3 Signal quality method for assessing 1D-DUS in a clinical environ-

ment 43

3.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.3.1 Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46

3.3.2 Segment selection . . . . . . . . . . . . . . . . . . . . . . . . . 47

3.3.3 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 48

3.3.4 Template-based quality assessment of 1-D Doppler Ultrasound 49

3.3.5 Sample Entropy and Power Spectrum Density (PSD) . . . . . 54

3.3.6 Feature Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.3.7 Classi_cation . . . . . . . . . . . . . . . . . . . . . . . . . . . 55

3.3.8 Method performance assessment . . . . . . . . . . . . . . . . . 56

3.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.4.1 Feature Selection . . . . . . . . . . . . . . . . . . . . . . . . . 58

3.4.2 Test Set Performance . . . . . . . . . . . . . . . . . . . . . . . 59

3.4.3 Performance of classi_er on intermediate quality segments . . 59

3.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60

3.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63

4 Data capture errors in the 1D-DUS recorded in the _eld 64

4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 64

4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65

4.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.3.1 Data acquisition . . . . . . . . . . . . . . . . . . . . . . . . . . 69

4.3.2 Description of data . . . . . . . . . . . . . . . . . . . . . . . . 70

4.3.3 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 70

4.3.4 Class annotation . . . . . . . . . . . . . . . . . . . . . . . . . 71

4.3.5 Feature extraction . . . . . . . . . . . . . . . . . . . . . . . . 73

4.3.6 Feature selection . . . . . . . . . . . . . . . . . . . . . . . . . 80

4.3.7 Classi_cation . . . . . . . . . . . . . . . . . . . . . . . . . . . 81

4.3.8 Performance assessment . . . . . . . . . . . . . . . . . . . . . 81

4.3.9 Processing time . . . . . . . . . . . . . . . . . . . . . . . . . . 84

4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 85

4.4.1 Model performance . . . . . . . . . . . . . . . . . . . . . . . . 85

4.4.2 Feature selection and real time performance . . . . . . . . . . 87

4.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 89

4.6 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91

5 Fetal heart estimation from 1D-DUS signal 92

5.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 92

5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93

5.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5.3.1 Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 96

5.3.2 Manual Heart Rate Estimation . . . . . . . . . . . . . . . . . 98

5.3.3 Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 103

5.3.4 Heart rate estimator . . . . . . . . . . . . . . . . . . . . . . . 104

5.3.5 Performance assessment . . . . . . . . . . . . . . . . . . . . . 111

5.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114

5.4.1 Optimized parameters and validation stage . . . . . . . . . . . 114

5.4.2 Oxford dataset . . . . . . . . . . . . . . . . . . . . . . . . . . 115

5.4.3 Guatemala RCT dataset . . . . . . . . . . . . . . . . . . . . . 116

5.4.4 Comparison between Oxford JR and Guatemala RCT datasets 118

5.5 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119

5.5.1 Interpretations of Findings . . . . . . . . . . . . . . . . . . . . 119

5.5.2 Study Limitations . . . . . . . . . . . . . . . . . . . . . . . . . 121

5.5.3 Future Directions . . . . . . . . . . . . . . . . . . . . . . . . . 123

5.6 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

6 Estimating birth weight from postnatal weights 124

6.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 124

6.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

6.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127

6.3.1 Infant weight development . . . . . . . . . . . . . . . . . . . . 127

6.3.2 Weight curve models . . . . . . . . . . . . . . . . . . . . . . 128

6.3.3 Birth weight in Guatemala . . . . . . . . . . . . . . . . . . . 130

6.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

6.4.1 Database . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 130

6.4.2 Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 131

6.4.3 Fitting models . . . . . . . . . . . . . . . . . . . . . . . . . . 132

6.4.4 Comparison with other models . . . . . . . . . . . . . . . . . 134

6.4.5 Estimating birth weight . . . . . . . . . . . . . . . . . . . . . 134

6.4.6 Comparison of the estimated birth weight . . . . . . . . . . . 135

6.4.7 Identi_cation of low birth weight . . . . . . . . . . . . . . . . 135

6.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 135

6.5.1 Preprocessed data . . . . . . . . . . . . . . . . . . . . . . . . . 135

6.5.2 Fitted models . . . . . . . . . . . . . . . . . . . . . . . . . . . 136

6.5.3 Comparison to other models . . . . . . . . . . . . . . . . . . . 137

6.5.4 Estimation of birth weight and identi_cation of low birth weight 137

6.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 140

6.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

7 Estimating gestational age from 1D-DUS and maternal blood pres-

sure recordings 144

7.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

7.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 145

7.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 147

7.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

7.4.1 Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150

7.4.2 Deriving the FHR signal . . . . . . . . . . . . . . . . . . . . . 153

7.4.3 Features used for gestational age estimation . . . . . . . . . . 155

7.4.4 Estimation of gestational age . . . . . . . . . . . . . . . . . . 161

7.4.5 Detecting possibles cases of IUGR . . . . . . . . . . . . . . . . 165

7.5 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

7.5.1 Training/Validation performance . . . . . . . . . . . . . . . . 166

7.5.2 Ranking the features . . . . . . . . . . . . . . . . . . . . . . . 167

7.5.3 Testing performance . . . . . . . . . . . . . . . . . . . . . . . 168

7.5.4 Comparing features for the estimation of GA . . . . . . . . . . 170

7.5.5 GA estimation errors as a function of birth weight . . . . . . . 171

7.6 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

7.6.1 Interpretations of Findings . . . . . . . . . . . . . . . . . . . . 172

7.6.2 Study Limitations . . . . . . . . . . . . . . . . . . . . . . . . . 174

7.6.3 Future directions . . . . . . . . . . . . . . . . . . . . . . . . . 176

7.7 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 176

8 Conclusion 177

8.1 Summary and contributions . . . . . . . . . . . . . . . . . . . . . . . 177

8.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 179

8.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 180

8.3.1 Online functionalities . . . . . . . . . . . . . . . . . . . . . . . 180

8.3.2 Segmentation of 1D-Doppler ultrasound signals into beat-tobeat

intervals . . . . . . . . . . . . . . . . . . . . . . . . . . . 181

8.3.3 Deep learning for improving gestational age estimations and

detection of IUGR cases . . . . . . . . . . . . . . . . . . . . . 182

Appendix A Table of all feature combinations for quality assessment 183

Appendix B Algorithm for determining periodicity 187

Bibliography 188

About this Dissertation

Rights statement
  • Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.
School
Department
Degree
Submission
Language
  • English
Research Field
Keyword
Committee Chair / Thesis Advisor
Committee Members
Last modified

Primary PDF

Supplemental Files