Due to the development of society, people's attention to health management and health pursuit has increased, and the need to monitor the health status of themselves or their family members has also increased.
At present, people use instruments to obtain vital information and monitor their health status. General monitoring instruments rely on contact sensors and electrodes for measurement, which requires direct or indirect contact with the human body, which has limited application scope and user compliance problems, and when users are unwilling to wear or forget to wear sensing equipment, these measurement equipment will not work normally.
In response to the above problems, the medical electronics team of the State Key Laboratory of Sensing Technology of the Aerospace Information Innovation Institute of the Chinese Academy of Sciences (CAA) is committed to the research of non-inductive medical and health monitoring based on millimeter-wave radar, which solves the problem of user compliance, avoids the discomfort caused by contact wearing equipment, and helps to integrate health monitoring into daily life.
At present, the typical applications of millimeter-wave radar in the medical and health field can be summarized into three categories: physiological signs monitoring, fall detection, human-computer interaction, etc.
1. Monitoring of physiological signs
The research team has developed an integrated system based on frequency modulated continuous wave (FMCW) radar to carry out continuous dynamic monitoring of non-inductive heart rate and blood pressure.
1.Heart rate monitoring
The research team proposed a sensitive human motion detection algorithm, an optimal distance element selection algorithm, and a global optimization model, which can identify the body movement state and maintain the accuracy of heart rate measurement in different individuals and different sleeping positions. In 91The median interval between beats (IBI) error was 12 ms and the median beat-by-beat heart rate (HR) error was 0 at 2% time coverage65bpm (Figure 1), the accuracy can reach the national standard of medical-grade heart rate monitoring equipment (<5bpm), and its accuracy is the highest under the same experimental conditions.
It has high accuracy and low computational complexity, and can be implemented on low-cost radar chips, which also provides the possibility for millimeter-wave radar health monitoring to enter life scenarios.
Figure 1 Comprehensive performance evaluation of heart rate calculation.
2.Blood pressure monitoring
For the first time, the research team proposed and developed a single-radar non-contact continuous blood pressure measurement system based on pulse wave transmission time, which uses single-millimeter-wave radar to realize real-time measurement of blood transmission time from the left ventricle of the heart to the carotid artery. The correlation coefficient between the pulse transmission time measured by the system and the transmission time measured by the wearable can reach 091, and its systolic blood pressure (SBP) and diastolic blood pressure (DBP) errors are 554±7.62 mmHg and 468±6.15 mmHg (Figure 2), which is close to the International Standard for the Evaluation of Electronic Blood Pressure Monitors (AAMI) issued by the American Association for the Advancement of Medical Devices, is therefore expected to provide continuous blood pressure measurement in a non-contact manner without interfering with the user's daily activities.
Fig.2 Correlation analysis of radar blood pressure monitoring.
2. Fall detection
In terms of fall detection, the team proposed a fall detection method based on multi-modal radar information fusion, which can obtain different radar feature information to complete the detection, which is helpful to help people obtain fall information in time and facilitate rescue.
For the first time, the model fuses information such as distance, velocity, azimuth angle, and pitch angle at the same time to achieve high-accuracy fall detection (Fig. 3), which can effectively distinguish between 52 daily non-fall actions and 12 fall actions, reaching 98 in the test of new users and new environments3% true positive rate and 005% false positive rate.
Fig.3 Fall detection frame of millimeter-wave radar.
In addition, the research team also proposed a fall detection model based on the idea of anomaly detection, which uses difficult sample mining techniques to reduce the false positive rate (Fig. 4), which can maintain high accuracy in complex real-world scenarios. In the case of training the model without using fall samples and without using the label information of non-fall samples, 9554% true positive rate and 107% false positive rate.
Fig.4 Unsupervised fall detection framework.
3. Human-computer interaction
The research team has developed a variety of models for human-computer interaction. For example, the open-set identification model based on cardiac radar signals, the gait recognition model based on semi-supervised learning Xi, and the semi-supervised gesture recognition framework. These models effectively facilitate the exchange of information between users and computer systems, and can realize personalized monitoring and management of users.
Based on the identity recognition model of cardiac radar signals, for the first time, the feasibility of using radar heartbeat signals for identity recognition under the assumption of open set is **, and the algorithm shows good effectiveness in both open and closed set environments, and the accuracy reaches 99 under the setting of closed set and open set, respectively17% and 9357%。
The gait recognition model based on semi-supervised learning Xi is the first radar-based semi-supervised gait recognition method. The co-training of radar signals in two modes greatly reduces the number of labeled samples required to train the model, which is conducive to the popularization and use of this technology. The method achieves a gait recognition accuracy of 90 using only 4 minutes of labeled samples per user7%。
The semi-supervised gesture recognition framework combines models and specific data augmentation techniques to make full use of large amounts of unlabeled mmWave gesture data to achieve high-accuracy gesture recognition. Under the setting of a cross-position domain with a training test ratio close to 1:4 and a cross-environment domain with a training test ratio close to 1:8, the gesture recognition accuracy of the model reached 98., respectively32% and 9739%。
At present, the non-contact medical and health monitoring system based on millimeter-wave radar has been clinically verified in some hospitals, and has broad application prospects in the field of medical care, intelligent elderly care, arrhythmia, stroke, chronic obstructive pulmonary disease and sleep apnea management in the future.
The above research results have been published in the IEEE Internet of Things Journal, Expert Systems with Applications and other journals of the Chinese Academy of Sciences.
Published**:
1.Fang Zhen, Jian Pu, Zhang Hao, et al. Review of non-contact medical and health monitoring technology based on FMCW radar[J].Journal of Radars, 2022, 11(3): 499-516 doi: 10.12000/jr22019.
2. zhang h, jian p, yao y, et al. radar-beat: contactless beat-by-beat heart rate monitoring for life scenes[j]. biomedical signal processing and control, 2023, 86: 105360.
3. geng f, bai z, zhang h, et al. contactless and continuous blood pressure measurement according to captt obtained from millimeter w**e radar[j]. measurement, 2023: 113151.
4. yao y, liu c, zhang h, et al. fall detection system using millimeter-w**e radar based on neural network and information fusion[j]. ieee internet of things journal, 2022, 9(21): 21038-21050.
5. yao y, zhang h, liu c, et al. unsupervised learning-based unobtrusive fall detection using fmcw radar[j]. ieee internet of things journal, 2023. doi: 10.1109/jiot.2023.3301887.
6. yan b, zhang h, yao y, et al. heart signatures: open-set person identification based on cardiac radar signals[j]. biomedical signal processing and control, 2022, 72: 103306.
7. yao y, zhang h, xia p, et al. mmsignature: semi-supervised human identification system based on millimeter w**e radar[j]. engineering applications of artificial intelligence, 2023, 126: 106939.
8. yan b, wang p, du l, et al. mmgesture: semi-supervised gesture recognition system using mmw**e radar[j]. expert systems with applications, 2023, 213: 119042.
**: Aerospace Information Research Institute, Chinese Academy of Sciences
The official account of the Voice of the Chinese Academy of Sciences is a 100-school assistance plan