An Underscreen Fingerprint Sensor for Measurement of PPG and Blood Pressure Estimation

• Published in: IEEE sensors journal Vol. 25; no. 7; pp. 10964 - 10976
• Main Authors: Nguyen, Duc Huy, Chao, Paul C.-P., Ngo, Duc Thang, Huang, Sheng-Chieh, Huang, Yen-Tang, Huang, Mei-Lien, Chen, Yan-Liang, Huang, Teng-Wei
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.04.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: 1-D convolutional neural network (1D-CNN); Accuracy; Algorithms; Artificial neural networks; Blood pressure; blood pressure (BP); Estimation; Fingerprint recognition; fingerprint sensor; Fingerprint verification; Long short term memory; long short-term memory (LSTM); Low noise; Machine learning; Photodiodes; photoplethysmography (PPG); Pixels; Predictive models; quality assessment (QA); Semiconductor devices; Sensor arrays; Sensors; Signal to noise ratio; Thin film transistors
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2025.3541447

A fingerprint sensor array of pixels with photodiodes for sensing and thin-film transistors (TFTs) as switches were developed for the first time together with new algorithms to estimate blood pressure (BP) based on quality photoplethysmography (PPG) sensed from this array device. Toward obtaining PPG in high quality, i.e., low noise, the pixels in the sensor array for underscreen fingerprint sensing in mobile devices are designed capable of being operated in a different mode, connected in parallel to output larger current signals than the original mode of fingerprint sensing, for better signal-to-noise ratio (SNR) of sensed PPGs. The sensed PPG is next preprocessed by being filtered to remove noise, normalized, resampled, and so on, followed by a quality check engineered to assess further the sensed PPG in real time by a built deep-learning (DL) model, a 1-D convolutional neural network (1D-CNN) with long short-term memory (LSTM). The model is intended to disqualify the PPGs contaminated by ambient light interference and motion artifacts. Data were collected from 88 subjects, including 52 men and 36 women, by measuring PPG signals from a fingerprint sensor, with systolic BP (SBP) ranging from 90 to 150 mmHg and diastolic BP (DBP) from 50 to 85 mmHg. An OMRON HEM-7127 device was used to obtain reference BPs. In the results, an accuracy of 97.13% for distinguishing between qualified and unqualified PPGs has been achieved by the aforementioned 1D-CNN trained by ground truths, precollected PPGs judged and labeled into two groups, qualified and unqualified PPGs. Next, only the qualified PPGs are fed to another DL model built for estimating BP, in a structure of 1D-CNN with three consecutive sections of convolution, batch normalization, and max pooling. This BP model is next optimized to a structure and parameters to achieve the best-combined estimation of SBP and DBP. The results show accuracy with 1.31 mean absolute error (MAE) <inline-formula> <tex-math notation="LaTeX">\pm ~1.34 </tex-math></inline-formula> standard deviation (SD) for SBP and 4.11 (MAE) <inline-formula> <tex-math notation="LaTeX">\pm ~3.32 </tex-math></inline-formula> (SD) for DBP, favorable as compared to all the prior arts using an image sensor to estimate BPs.