CNN-Based LCD Transcription of Blood Pressure From a Mobile Phone Camera

Kulkarni, Samruddhi Shreeram; Katebi, Nasim; Valderrama, Camilo E.; Rohloff, Peter; Clifford, Gari D.

doi:10.3389/frai.2021.543176

Cited by 8 publications

(6 citation statements)

References 41 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Physiology or vitals: Through photographs of medical devices, vital signs such as blood pressure can be logged at the time of collection. Future work will focus on autotranscribing readings, as demonstrated in [ 36 ].…”

Section: Resultsmentioning

confidence: 99%

An Open-Source Privacy-Preserving Large-Scale Mobile Framework for Cardiovascular Health Monitoring and Intervention Planning With an Urban African American Population of Young Adults: User-Centered Design Approach

Clifford¹,

Nguyen²,

Shaw³

et al. 2022

JMIR Form Res

View full text Add to dashboard Cite

Background Cardiovascular diseases (CVDs) are the leading cause of death worldwide and are increasingly affecting younger populations, particularly African Americans in the southern United States. Access to preventive and therapeutic services, biological factors, and social determinants of health (ie, structural racism, resource limitation, residential segregation, and discriminatory practices) all combine to exacerbate health inequities and their resultant disparities in morbidity and mortality. These factors manifest early in life and have been shown to impact health trajectories into adulthood. Early detection of and intervention in emerging risk offers the best hope for preventing race-based differences in adult diseases. However, young-adult populations are notoriously difficult to recruit and retain, often because of a lack of knowledge of personal risk and a low level of concern for long-term health outcomes. Objective This study aims to develop a system design for the MOYO mobile platform. Further, we seek to addresses the challenge of primordial prevention in a young, at-risk population (ie, Southern-urban African Americans). Methods Urban African Americans, aged 18 to 29 years (n=505), participated in a series of co-design sessions to develop MOYO prototypes (ie, HealthTech Events). During the sessions, participants were orientated to the issues of CVD risk health disparities and then tasked with wireframing prototype screens depicting app features that they considered desirable. All 297 prototype screens were subsequently analyzed using NVivo 12 (QSR International), a qualitative analysis software. Using the grounded theory approach, an open-coding method was applied to a subset of data, approximately 20% (5/25), or 5 complete prototypes, to identify the dominant themes among the prototypes. To ensure intercoder reliability, 2 research team members analyzed the same subset of data. Results Overall, 9 dominant design requirements emerged from the qualitative analysis: customization, incentive motivation, social engagement, awareness, education, or recommendations, behavior tracking, location services, access to health professionals, data user agreements, and health assessment. This led to the development of a cross-platform app through an agile design process to collect standardized health surveys, narratives, geolocated pollution, weather, food desert exposure data, physical activity, social networks, and physiology through point-of-care devices. A Health Insurance Portability and Accountability Act–compliant cloud infrastructure was developed to collect, process, and review data, as well as generate alerts to allow automated signal processing and machine learning on the data to produce critical alerts. Integration with wearables and electronic health records via fast health care interoperability resources was implemented. Conclusions The MOYO mobile platform provides a comprehensive health and exposure monitoring system that allows for a broad range of compliance, from passive background monitoring to active self-reporting. These study findings support the notion that African Americans should be meaningfully involved in designing technologies that are developed to improve CVD outcomes in African American communities.

show abstract

Section: Resultsmentioning

confidence: 99%

An Open-Source Privacy-Preserving Large-Scale Mobile Framework for Cardiovascular Health Monitoring and Intervention Planning With an Urban African American Population of Young Adults: User-Centered Design Approach

Clifford¹,

Nguyen²,

Shaw³

et al. 2022

JMIR Form Res

View full text Add to dashboard Cite

show abstract

“…Hence, to increase the adoption of the PTEC HAT Programme in the subsequent implementation scaling phase, it is important to address this barrier and make the transmission of readings effortless and easy. One solution for the implementation scaling phase would be for the study team to explore alternative modes of transferring their readings, for example, via optical character reading (OCR) or manual submission [ 27 ].…”

Section: Discussionmentioning

confidence: 99%

Primary Technology Enhanced Care Home HbA1c Testing (PTEC HAT) programme: a feasibility pilot study in Singapore

Tyagi,

Koh,

Lee

et al. 2024

BMC Prim. Care

View full text Add to dashboard Cite

Background Considering time-consuming, cost-related limitations of laboratory-based HbA1c testing and follow-up clinic visits for diabetes management, it is important to explore alternative care models which incorporate point-of-care testing for HbA1c to monitor glycaemic control and related management. Methods Therefore, we adopted an implementation perspective to conduct one group pre- and post-intervention feasibility pilot assessing feasibility, acceptability and satisfaction with conducting home HbA1c test by patients with type 2 diabetes coupled with telemonitoring and teleconsultations (i.e., the Primary Technology Enhanced Care (PTEC) Home HbA1c Testing (HAT) Programme) in Singaporean primary care setting. The secondary objective was to compare the HbA1c, blood pressure and primary care visits at the end or during intervention, vs. 6 months before. Adult patients with type 2 diabetes with HbA1c ≤ 8% without any diabetes complications and having phone compatibility were recruited. Data was collected via patient self-reports and electronic medical records extraction. While summary statistics and paired t-test were computed for quantitative data, open-ended feedback was analysed using content analysis. Results A total of 33 participants completed the intervention out of 37 (33/37 = 89%) recruited from 73 eligible (37/73 = 51%). Most were either 51 to 60 years old (46.9%) or more than 60 years (37.5%), with more males (53.1%) and majority Chinese (93.8%). Majority (81.3%) felt that home HbA1c testing was beneficial with most commonly reported benefit of not having a clinic visit. A key finding was the average of diabetes-related visits being significantly lower post-intervention with comparable HbA1c values pre- and post-intervention. The most commonly reported challenge was using Bluetooth to transmit the reading (43.7%), followed by having too many steps to remember (28.1%). While participants reported being overall satisfied with the intervention, only 22% were willing to pay for it. Conclusion Our findings support home HbA1c testing by patients coupled with telemonitoring and teleconsultations. Following are practical recommendations for the implementation scaling phase: offering PTEC HAT Programme to suitable patients who are self-motivated and have adequate digital literacy, provision of adequate educational and training support, sending reminders and exploring enabling manual submission of HbA1c readings considering Bluetooth-related challenges.

show abstract

“…In earlier work, we showed that deep-learning-based digit recognition from photos of medical device displays can help accurately capture such data [20]…”

Section: Background Mhealth System For Bp Monitoringmentioning

confidence: 99%

“…An additional "blank" character was incorporated for shorter sequences. In this work, the CNN-based model [20] was used to detect the sequences of digits. In this model a softmax classifier, receives extracted features from X by a convolutional neural network and returns the probability of each digit (Fig 5).…”

Section: Digit Recognitionmentioning

confidence: 99%

See 1 more Smart Citation

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

Katebi,

Bremer,

Nguyen

et al. 2023

Preprint

Self Cite

View full text Add to dashboard Cite

This paper introduces a novel approach to address the challenges associated with transferring blood pressure (BP) data from oscillometric devices used in self-measured BP monitoring systems. The primary objective of this study is to improve the accessibility and usability of BP data for monitoring and managing BP during pregnancy and postpartum, particularly in low-resource settings. To this end, we developed an automated image transcription technique to effectively transcribe readings from BP devices. The photos of the BP devices were captured as part of perinatal mobile health (mHealth) monitoring systems, conducted in four studies across two countries. The Guatemala Set 1 and Guatemala Set 2 datasets include the data captured by a cohort of 49 lay midwives from 1697 and 584 pregnant women carrying singletons in the second and third trimesters in rural Guatemala during routine screening. Additionally, we designed an mHealth system in Georgia for postpartum women to monitor and report their BP at home with 23 and 49 African American participants contributing to the Georgia I3 and Georgia IMPROVE projects, respectively. We developed a deep learning-based model which operates in two steps: LCD localization using the You Only Look Once (YOLO) object detection model and digit recognition using a convolutional neural network-based model capable of recognizing multiple digits. We applied color correction and thresholding techniques to minimize the impact of reflection and artifacts. Three experiments were conducted based on the devices used for training the digit recognition model. Overall, our results demonstrate that the device-specific model with transfer learning and the device independent model outperformed the device-specific model without transfer learning. The mean absolute error (MAE) of image transcription on held-out test datasets using the device-independent digit recognition were 1.1 and 1.1 mmHg for systolic and diastolic BP in the Georgia IMPROVE and 1 and 0.6 mmHg in Guatemala Set 2 datasets. The MAE, far below the FDA requirement of 5 mmHg, makes the proposed model suitable for general use when used with appropriate error devices.

show abstract

CNN-Based LCD Transcription of Blood Pressure From a Mobile Phone Camera

Cited by 8 publications

References 41 publications

An Open-Source Privacy-Preserving Large-Scale Mobile Framework for Cardiovascular Health Monitoring and Intervention Planning With an Urban African American Population of Young Adults: User-Centered Design Approach

An Open-Source Privacy-Preserving Large-Scale Mobile Framework for Cardiovascular Health Monitoring and Intervention Planning With an Urban African American Population of Young Adults: User-Centered Design Approach

Primary Technology Enhanced Care Home HbA1c Testing (PTEC HAT) programme: a feasibility pilot study in Singapore

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

Contact Info

Product

Resources

About