2023 Fiscal Year Research-status Report
Machine Learning Enabled Non-contact Sensing Platform for Blood Pressure and Glucose Prediction
| Project/Area Number |
23K11341
|
|---|
| Research Institution | Waseda University |
|---|
Principal Investigator |
劉 江 早稲田大学, 理工学術院, 教授 (50546851)
|
|---|
| Co-Investigator(Kenkyū-buntansha) |
中田 孝明 千葉大学, 大学院医学研究院, 教授 (20375794)
嶋本 薫 早稲田大学, 理工学術院, 教授 (80235639)
|
|---|
| Project Period (FY) |
2023-04-01 – 2026-03-31
|
| Keywords | Non-contact Sensing / Blood Pressure / Glucose / Machine Learning |
|---|
| Outline of Annual Research Achievements |
The demand for comfortable healthcare services continues to increase among patients and the public. To meet the needs of healthcare products, this project aims to design and develop a contactless sensing platform for predicting blood pressure and glucose levels. This academic year, we have proposed a non-contact method for estimating blood pressure using 2.4GHz microwave signals. The microwave signals are reflected from the body, and the time-varying reflection intensity was acquired during the process. The experiment begins with estimating the pulse waveform through post-processing of the acquired raw data. Then, blood pressure was estimated via machine learning methods with feature values derived from the detected pulse waveform.
Regarding glucose-sensing research, we have investigated the absorbance of each wavelength using an FTIR (Fourier Transform Infrared Spectroscopy) device. We found that mid-infrared light is sensitive to glucose levels. Based on these observations, we designed experimental circuits with lower costs compared to FTIR.
Part of these results were presented at the IEEE International Conference CCNC 2024 and the Global Information and Telecommunication Workshop 2023.
|
| Current Status of Research Progress |
Current Status of Research Progress
2: Research has progressed on the whole more than it was originally planned.
Reason
To estimate blood pressure, a total of 16 feature values were utilized. The heart rate was calculated as the mean of the results obtained from the chest and wrist. The performance of our estimation was evaluated using five 5 categories: Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), R^2 score, Standard deviation of Absolute Error (SDAE), and Correlation Coefficient (CC). The results indicated that the estimated values achieved a MAE within ±5 mmHg and a SDAE within 8mmHg for the Systolic blood pressure (SBP) and Diastolic blood pressure (DBP), respectively.
For Glucose estimation, a self-made infrared experimental instrument with a peak wavelength of 2900 nm was designed to obtain the physical signal. Preliminary experiments have been conducted, and the results indicate that this research is progressing in the right direction. We believe this system has significant potential for the future intelligent health monitoring systems.
|
| Strategy for Future Research Activity |
For blood pressure prediction, in our previous progress, the predicted values met the accuracy standards defined by the Japanese Industrial Standards (JIS). However, multiple steps were required to estimate blood pressure. In the next phase, we plan to create a user-friendly blood pressure monitoring system that can be integrated into smart devices such as smartphones.
Regarding glucose prediction, we have investigated the absorbance of each wavelength using an FTIR device. We found that mid-infrared light is sensitive to glucose levels. A self-made infrared experimental instrument using LEDs with the appropriate wavelength will be designed to obtain the physical signal. Experiments with human subjects will be conducted, and machine learning algorithms will be used for the blood glucose system estimation.
|
| Causes of Carryover |
We ordered the experimental equipment (2900 nm LEDs and photodiodes), but due to delivery delays, we decided to pay for it after delivery using next year's research budget.
|
