Automated Image Transcription for Perinatal Blood Pressure Monitoring using Mobile Health Technology

Nasim Katebi, Whitney Bremer, Tony Nguyen, Daniel Phan, Jamila Jeff, Kirkland Armstrong, Paula Phabian-Millbrook, Marissa Platner, Kimberly Carroll, Banafsheh Shoai, Peter Rohloff, Sheree L. Boulet, Cheryl G. Franklin, Gari D. Clifford

Abstract

This paper introduces a novel approach to address the challenges associated with transferring blood pressure (BP) data obtained from oscillometric devices used in self-measured BP monitoring systems to integrate this data into medical health records or a proxy database accessible by clinicians, particularly in low literacy populations. 

To this end, we developed an automated image transcription technique to effectively transcribe readings from BP devices, ultimately enhancing the accessibility and usability of BP data for monitoring and managing BP during pregnancy and the postpartum period, particularly in low-resource settings and low-literate populations. 

In the designed study, the photos of the BP devices were captured as part of perinatal mobile health (mHealth) monitoring programs, conducted in four studies across two countries.

Introduction

Hypertensive disorders of pregnancy (HDP) are the most common medical complication encountered during pregnancy [1]. HDPs are related to a combination of maternal, placental and fetal factors and can lead to serious complications which can cause maternal and fetal morbidity and mortality [2]. The burden of these complications is disproportionately borne by women in low and middle-income countries (LMICs) and resource-constrained areas of high-income countries. 

For example, in Latin America, pregnancy vascular disorders are the leading cause of maternal mortality where up to 26% of maternal deaths are estimated to be related to preeclampsia [3, 4]. In the USA, during 2017–2019, the prevalence of HDP among delivery hospitalizations increased from 13.3% to 15.9% [5]. This trend is particularly concerning given the existing disparities in maternal health outcomes across different regions.

Background

mHealth system for BP monitoring

mHealth BP monitoring systems have demonstrated superior performance in comparison to traditional methods of BP monitoring, particularly in terms of convenience and management of hypertension [19, 20]. In these monitoring systems, once BP data has been measured, several methods allow users to record and transmit this data to clinicians. Core elements of number digitization are manual transcription on both paper and smartphone, Bluetooth or cellular data receivers and memory-card based and USB transfer. Each of these approaches has potential benefits and drawbacks, particularly in terms of risk of missing and inaccurate data. In the manual transcription, users may introduce errors during the transfer of data from the device display.

Method

In this section the step-by-step approach to convert BP images into numerical format is described including LCD localization and the digit recognition methods.

Automatic LCD localization

Accurately localizing the LCD frames is essential for converting the images into a numerical format, but this can be challenging due to orientation and zooming effects, resulting in differences in the size and location of the frames. Over the past decade, rapid advancements in deep learning have driven extensive research and significant contributions aimed at improving the performance of object detection.

Results

We compared two different LCD localization methods and their impact on BP transcription accuracy. The YOLO-based method and the contour-based method were tested on the same set of data previously used in a study by Kulkarni et al. [17]. Our results show that the YOLO-based method outperformed the contour-based LCD localization method as shown in Table 2. In this experiment, the model was trained on 5020 single LCD images and tested on 1677 images. The results of BP transcription, showed that the YOLO-based method improved both the accuracy and MAE of transcribing SBP and DBP.

Citation: Katebi N, Bremer W, Nguyen T, Phan D, Jeff J, Armstrong K, et al. (2024) Automated image transcription for perinatal blood pressure monitoring using mobile health technology. PLOS Digit Health 3(10): e0000588. https://doi.org/10.1371/journal.pdig.0000588

Editor: Esli Osmanlliu, McGill University, CANADA

Received: June 11, 2023; Accepted: July 22, 2024; Published: October 2, 2024

Copyright: © 2024 Katebi et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: Images used in this research include reflections of participants faces and therefore constitute protected personal information and cannot be posted publicly. Data access for research will require appropriate IRB and HIPAA-compliant security documentation, and the signing of an institutional data use agreement. The institutional email for requests is: datarequests@dbmi.emory.edu.

Funding: Research reported in this publication was supported in part by the National Institutes of Health, through the Fogarty International Center and the Eunice Kennedy Shriver National Institute of Child Health and Human Development (NICHD), grant number 1R21HD084114-01 to GC (Mobile Health Intervention to Improve Perinatal Continuum of Care in Guatemala), NICHD grant number 1R01HD110480 to GC (AI-driven low-cost ultrasound for automated quantification of hypertension, preeclampsia, and IUGR), Google.org AI for the Global Goals Impact Challenge Award to GC, NK and PR, the Imagine, Innovate and Impact (I3) Funds from the Emory School of Medicine to GC and SB and through the Georgia IMPROVE on Maternal Health, funded by NIH National Center for Advancing Translational Sciences (NCATS) as an Administrative Supplement to the Georgia Clinical and Translational Alliance, grant number UL1-TR002378 to GC, SB and CF. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health. NK is partially funded by a PREHS-SEED award grant K12ESO33593. We would also like to acknowledge the support of the Grady Health System, Atlanta, Georgia in conducting this research. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.