A 24-hour BP estimation algorithm using wrist PPG signals from smart watch is developed. A signal quality indicator is established to assess the applicability of BP estimation during daily life condition. A personal model based on deep neural network is constructed by model agnostic meta learning (MAML) technique so that few training data are required. The average root mean square error is 7.65 mmHg. A promising solution for long-termpervasive healthcare monitoring thus can be provided.

A 24-hour personal BP estimation algorithm based on model agnostic meta learning technique is developed so that few training data are required. The applicability of BP estimation in daily life is assessed by the skewness of wrist PPG signalsthe activity count from 3-axis accelerometer,thus negligible interference is generated to the user. From experiments of 14 subjects, the average RMSE is 7.65 mmHg, which can help to evaluate the extent of BP dippotential cardiovascular risk.

The technique is highly related with the ICT industry, the medical field,the healthcare industry. The enhanced capability of smart watch for healthcare can increase its add-on value. The continually-estimated BPs in daily life can provide diurnal BP trends for doctors to achieve more effective BP controlcardiovascular risk reduction. The long-termpervasive health monitoring can help users for disease preventionto strengthen the motivation for healthy life style adoption.