Detection of Adverse Events in Pregnancy Using a Low-Cost 1D Doppler Ultrasound Signal Open Access
Katebijahromi, Nasim (Summer 2021)
Abstract
Fetal maternal mortality is an enormous global health challenge, affecting over 2.6 million families annually. The burden is most heavily felt in low-and middle-income countries (LMICs) due to systemic healthcare issues related to inequity, limited funding for medical technology, and poor infrastructure for delivering and maintaining technology. Fetal growth restriction is increasingly recognized as an important contributor to fetal health problems in LMICs. One of the effective approaches to detect fetal developmental issues is tracking fetal heart rate variability (FHRV). FHRV is also an indicator of fetal brain development since it is influenced by the autonomic nervous system, which evolves during pregnancy. Therefor, accurate estimation of gestational age using FHRV patterns could help to detect fetal brain developmental issues and potential cases of small for gestational age. This thesis aims to enhance fetal health monitoring for disadvantaged populations by developing AI-enabled edge computing devices which are intuitive to use even for low-literacy populations. Specifically, this work presented machine learning methods to analyze one-dimension fetal Doppler ultrasound signals (1D-DUS), which have been collected using a low-cost mobile health system.
Developing accurate models to capture the underlying dynamic of 1D-DUS is a challenging task. Since 1D-DUS is nonstationary, highly susceptible to noise and movement, and has a transient nature. Using additional device for recording another data modality or labels of the beat intervals can mitigate the effect of highly variable morphology. However, these solutions significantly complicate the use and raise the cost of the smartphone-mediated perinatal screening system. Therefore, this thesis aimed to underline the importance of taking challenges in LMICs into account for developing accurate machine learning methods and addressed the challenges via two approaches: 1) developing an unsupervised probabilistic segmentation method to estimate FHRV metrics from 1D-DUS recordings. 2) Developing a deep sequence learning model with an attention mechanism for automatic feature extraction and estimation of fetal gestational age. This model is the first attempt to estimate gestational age from only Doppler signals and outperforms previous attempts based on multiple signals. The developed methods will ultimately run on-device and interact with the healthcare workers or mothers directly. This work could assist traditional birth attendants in rural areas with a decision support system to identify patients with possible pregnancy-related abnormalities for early triage and intervention.
Table of Contents
Contents
1 Introduction1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .1
1.2 Aim of this thesis . . . . . . . . . . . . . . . . . . . . . . . . . . . . .4
1.3 Thesis outline . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .5
1.4 List of publication . . . . . . . . . . . . . . . . . . . . . . . . . . . .6
2 A review of fetal cardiac monitoring, with a focus on low-and middle-income countries8
2.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .8
2.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .9
2.3 Fetal cardiac circulation . . . . . . . . . . . . . . . . . . . . . . . . .11
2.3.1Control of fetal heart rate . . . . . . . . . . . . . . . . . . . .11
2.4 Fetal heart monitoring techniques . . . . . . . . . . . . . . . . . . . .14
2.4.1Fetal phonocardiogram. . . . . . . . . . . . . . . . . . . . .14
2.4.2One-dimensional Doppler ultrasound . . . . . . . . . . . . . .15
2.4.3Cardiotocography . . . . . . . . . . . . . . . . . . . . . . . . .17
2.4.4Fetal electrocardiogram . . . . . . . . . . . . . . . . . . . . . .19
2.4.5Fetal magnetocardiography . . . . . . . . . . . . . . . . . . .21
2.5 Ultrasound imaging . . . . . . . . . . . . . . . . . . . . . . . . . . . .21
2.5.1Fetal biometry . . . . . . . . . . . . . . . . . . . . . . . . . .23
2.5.2Doppler velocimetry . . . . . . . . . . . . . . . . . . . . . . .24
2.5.3Fetal echocardiography . . . . . . . . . . . . . . . . . . . . . .26
2.6 Comparison of fetal cardiac monitoring methods . . . . . . . . . . . .27
2.6.1Cost analysis and availability in LMICs . . . . . . . . . . . . .27
2.7 Usage of devices in LMICs . . . . . . . . . . . . . . . . . . . . . . . .30
2.8 Telemonitoring for perinatal care, an alternative for LMICs . . . . . .32
2.9 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . .34
3 Data collection and labelling 37
3.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37
3.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .37
3.3 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39
3.3.1Databases . . . . . . . . . . . . . . . . . . . . . . . . . . . . .39
3.3.2Data labeling and annotation . . . . . . . . . . . . . . . . . .40
3.3.3Annotation of the Leipzig university hospital database . . . .42
3.3.4Annotation of Oxford JR database . . . . . . . . . . . . . . .45
3.4 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . .47
4 Unsupervised fetal Doppler signals segmentation and heart rate vari-ability estimation48
4.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .48
4.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .50
4.3 Background . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .51
4.3.1Clinical trials using FHRV . . . . . . . . . . . . . . . . . . . .51
4.3.2Fetal heart rate variability estimation from Doppler . . . . . .52
4.3.3Beat segmentation in the 1D DUS . . . . . . . . . . . . . . . .53
4.4 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .54
4.4.1Datasets . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .55
4.4.2Pre-processing . . . . . . . . . . . . . . . . . . . . . . . . . . .56
4.4.3Data Transformation . . . . . . . . . . . . . . . . . . . . . . .57
4.4.4Clustering . . . . . . . . . . . . . . . . . . . . . . . . . . . . .61
4.4.5Hidden Semi-Markov Model . . . . . . . . . . . . . . . . . . .62
4.5 Performance assessment . . . . . . . . . . . . . . . . . . . . . . . . .65
4.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .67
4.6.1Analysis of dataset 1 . . . . . . . . . . . . . . . . . . . . . . .67
4.6.2Analysis of dataset 2 . . . . . . . . . . . . . . . . . . . . . . .68
4.7 Discussion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .70
4.8 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .73
5 Deep sequence learning for gestational age estimation77
5.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .77
5.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .78
5.3 Gestational age estimation model . . . . . . . . . . . . . . . . . . . .80
5.4 Dataset . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .81
5.5 Data analysis and feature extraction . . . . . . . . . . . . . . . . . .81
5.5.1Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . .81
5.5.2Time-frequency (TF) features for DUS components . . . . . .82
5.5.3Sequence modeling . . . . . . . . . . . . . . . . . . . . . . . .83
5.6 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .84
5.7 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . .85
6 Detection of noisy gestational age recordings86
6.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .86
6.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .87
6.3 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .89
6.3.1Data acquisition . . . . . . . . . . . . . . . . . . . . . . . . .89
6.3.2Maternal healthcare assessment . . . . . . . . . . . . . . . . .90
6.3.3Multiple measurement fusion . . . . . . . . . . . . . . . . . .90
6.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .91
6.4.1Distribution of the visits . . . . . . . . . . . . . . . . . . . . .91
6.4.2Evaluation of gestational age labels . . . . . . . . . . . . . . .91
6.5 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . .93
7 Hierarchical attention network for gestational age estimation96
7.1 Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .96
7.2 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .97
7.3 Related Work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .99
7.4 Method . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 100
7.4.1Hierarchical attention network for modeling long- and short-term temporal patterns . . . . . . . . . . . . . . . . . . . . . . 101
7.4.2Sequence Encoder . . . . . . . . . . . . . . . . . . . . . . . . . 103
7.4.3Hirarchical Attention . . . . . . . . . . . . . . . . . . . . . . . 103
7.5 Experimental set-up . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.5.1Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
7.5.2Preprocessing . . . . . . . . . . . . . . . . . . . . . . . . . . . 105
7.5.3Network implementation . . . . . . . . . . . . . . . . . . . . . 105
7.6 Evaluation metrics . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.7 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
7.8 Discussion and conclusion . . . . . . . . . . . . . . . . . . . . . . . . 107
8 Discussion and conclusions109
8.1 Summary and contributions . . . . . . . . . . . . . . . . . . . . . . . 109
8.2 Limitations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
8.3 Future work . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 111
About this Dissertation
School | |
---|---|
Department | |
Degree | |
Submission | |
Language |
|
Research Field | |
Keyword | |
Committee Chair / Thesis Advisor | |
Committee Members |
Primary PDF
Thumbnail | Title | Date Uploaded | Actions |
---|---|---|---|
Detection of Adverse Events in Pregnancy Using a Low-Cost 1D Doppler Ultrasound Signal () | 2021-09-10 12:21:21 -0400 |
|
Supplemental Files
Thumbnail | Title | Date Uploaded | Actions |
---|