论文部分内容阅读
The diverse breakthroughs in wearable sensors technology coupled with advanced biomedical signal processing have made it possible to keep track of basic human vitals like heart rate, blood glucose and blood pressure, etc.More convenient self-monitoring solutions based on these sensors are now available that not only help living a healthy daily life but also plays an important role in disease diagnosis and prognosis.However, each user might have different requirements and priorities when it comes to selecting a self-monitoring solution.After extensive research and careful selection, non-invasive blood glucose monitoring methods from the recent five years (2012-2016) are reviewed and analyzed based on certain constraints which include time efficiency, comfort, cost, portability, power consumption etc.a user might experience when using such techniques.Re-calibration, time and power efficiency are the biggest challenges that require further research in order to satisfy a large number of users.In order to solve these challenges, Artificial intelligence (AI) is being employed by many researchers.AI based estimation and decision models hold the future of non-invasive selfmonitoring in terms of accuracy, cost effectiveness, portability and efficiency etc.Photoplethysmography (PPG) is one of the many optical technologies that are the basis of these wearable sensors.An energy efficient algorithm to estimate pulse rate has been proposed in this work that uses smartphone camera as a sensor to acquire PPG signals.The use of smartphone camera makes the system simple, cost effective and portable.The percentage accuracy of the proposed algorithm is found to be 98.02％.Later, PPG signals are used to estimate blood glucose levels by applying machine learning algorithms.The blood glucose scale is divided into 6 groups; G1-G6.The estimation model is based on 67 features of single period waveform shape and classifies the collected data into 7 groups i.e.invalid data (G0) and G1-G6.The glucose estimation model is found to classify the PPG signals into respective groups with an overall accuracy of 75％ whereas for G2 (70-100 mg/dl) and G3 (101-130 mg/dl), the combined accuracy was found to be 80％.The invalid single periods were classified with an accuracy of 86.9％.Due to the classification results for the blood glucose range 70-130 mg/dl, another estimation model was designed using the smartphone PPG.The data was acquired from the left index finger of the volunteers in the form of a video.The blood glucose levels of the individuals range from 70-130 mg/dl.First the single periods of the PPG signal were classified as valid and invalid, and later the valid periods were classified into two groups i.e.G1 (70-100 mg/dl) and G2 (101-130 mg/dl).The overall accuracy of the proposed model was found to be 86.2％ whereas the accuracy of the invalid single period classification was found to be 98.2％.