AUTHOR=Kulkarni Samruddhi S. , Katebi Nasim , Valderrama Camilo E. , Rohloff Peter , Clifford Gari D. TITLE=CNN-Based LCD Transcription of Blood Pressure From a Mobile Phone Camera JOURNAL=Frontiers in Artificial Intelligence VOLUME=Volume 4 - 2021 YEAR=2021 URL=https://www.frontiersin.org/journals/artificial-intelligence/articles/10.3389/frai.2021.543176 DOI=10.3389/frai.2021.543176 ISSN=2624-8212 ABSTRACT=Routine blood pressure (BP) measurement in pregnancy is commonly done using automated oscillometric devices. Since no wireless oscillometric BP device validated in preeclamptic populations exists, a simple approach to capture readings from such devices is needed, especially in low-resource settings where transmission of BP data from the field to central locations is an important mechanism for triage. A total of 8192 BP readings were captured from the Liquid Crystal Display(LCD) screen of a standard Omron M7 self-inflating BP cuff using a cellphone camera. A cohort of 49 lay midwives captured this data from 1697 pregnant women carrying singletons between 6 weeks and 40 weeks gestational age in rural Guatemala during routine screening. Images showed a wide variability in their appearance due to variations in orientation and parallax; environmental factors such as lighting, shadows; and image acquisition factors such as motion blur and focus problems. They were independently labelled for readability and quality by three annotators (BP range: 34 – 203 mmHg) and disagreements were resolved. In this study, the authors proposed an approach to preprocess and automatically segment image under test into diastolic BP, systolic BP and heart rate using a contour-based technique. A deep convolutional neural network was trained to convert the images of each reading into numerical values using a multi-digit recognition approach. On \textcolor{blue}{readable low and high quality} images, the approach achieved a \textcolor{blue}{91\%} classification accuracy and mean absolute error of 3.19 mmHg for systolic BP and \textcolor{blue}{91\%} accuracy and mean absolute error of 0.94 mmHg for diastolic BP. \textcolor{blue}{These error values are within the FDA guidelines for BP monitoring when poor quality images are excluded. The performance of the proposed approach was evaluated against that of open-source OCR tools like Tesseract and Google API, thereby giving superior results. The algorithm was developed such that it could be deployed on a phone and work without connectivity to a network.