Notes specific to vital data
This chapter explains the basics of analyzing vital data generated by non-medical wellness devices such as JINS MEME and fitness trackers.
Features of each sensor
If you are considering using JINS MEME data(EOG), you may find that you can obtain indicators that cannot be obtained with other devices, but compared to other data, each has the characteristics shown in the table below.
Indicator | Acquisition Device | Signal Intensity | Signal Characteristics | Variation Period | Reflected Biological Phenomena | Acquisition Rate(Static) | Acquisition Rate(Active) |
---|---|---|---|---|---|---|---|
Heart rate (HR) | LED + light intensity sensor | Weak | periodic | 5-10 seconds | exercise intensity, tension, etc. | 95% | 70% |
HRV(HRV,PPI,LF/HF) | LED+light sensor | Weak | State | 5-10 sec | tension, stress, etc. | 80% | 0% |
HRV(HRV,RRI,LF/HF) | Amplifier + Potential Sensor | Medium | State | 5-10 sec | Tension, Stress, etc. | 99% | 85% |
ECG | Amplifier + Potential Sensor | Medium | State | 0.01-1 sec (within 1 beat) | Determination of Heart Disease | 99% | 0-10% |
Pedometer | Accelerometer | Strong | state | 0.3-0.5 sec | Number of steps, calories burned | 99% | 99% |
Activity intensity | Accelerometer | Strong | State | 1-2 seconds | Activity, calories expended | 99% | 99% |
EOG | Amplifier + Potential sensor | Weak | State | 1-2 seconds | Concentration, arousal, etc. | 90-95% | 0-10% |
EEG | Amplifier + Potential Sensor | Very weak | State | 1-2 seconds | Relative brain activity evaluation | 70% | 0% |
*These are reference values for acquisition rates.
The familiar indicators are the heart rate and pedometer, both of which have a characteristic that their accuracy hardly drops even static or active, and this is due to either of the following features.
- The signal itself is strong, so the data is not blurred even if there is some noise (pedometer)
- The signal itself is weak, but the signal is periodic, so data loss can be recovered through approaches such as Fourier transform (heart rate)
For example, LED heart rate variability measurement uses the same mechanism as heart rate but misses peaks due to slight noise, making stable measurement difficult. This is a more difficult task.
EEG is commonly used to monitor mental status, but it is not an easy sensor to use because it takes time to adjust before starting measurement and the signal is very weak and is canceled by EMG and EOG depending on the position of the sensor.
EOG is easy to use to obtain state of mental activity at rest. However, even at rest, there are durations where data cannot be obtained due to reapplication of glasses, chewing, etc. Therefore, it is necessary to assume a certain amount of missing data when analysis.
JINS MEME Indicators and Robustness
The following table shows the intermediate indices acquired by the EOG and their robustness.
Intermediate indices | Indicator sample | Robustness(1: Weak….5: Strong) |
---|---|---|
Average blink strength | Tension, arousal | 3 |
Standard deviation of blink strength | Stability | 3 |
Average blink speed* | Weak arousal/Sleepiness | 3 |
Standard deviation of blink speed | – | 3 |
Mean of blink frequency and interval | Immersion | 2 |
Number of eye movements (vertical) | Vitality | 1 |
Number of eye movement (horizontal) | Vitality, attention | 4 |
*A simple average is currently used for the blink interval, but may be switched to a weighted average in the future.
The blink frequency and interval data is greatly affected when data is missing due to noise, etc. Although JINS MEME data is processed to leave only clean intervals, the robustness of the data is lower than other data. In addition, blink and eye movement (vertical) judgments may be less accurate due to (1) the electrode between the eyebrows floating depending on the state of spectacle wear and (2) false negative and false positive judgments between blink and eye movement (vertical) because the vertical signal is classified.
The following table shows the indicators acquired by the 6-axis motion sensor and their robustness.
Indicator name | Robustness(1: Weak….5: Strong) |
---|---|
Walking Vibration | 4 |
Head movement (non-walking) | 4 |
Maximal Landing Strength | 4 |
Left/Right Foot Judgment | 2 |
Average of Posture Angle | 5 |
Standard deviation of posture angle | 5 |
Data from the 6-axis motion sensor can basically be used full time and can be used as is through a filter that confirms minimal wear.
Difficulty of building a new analytical framework
Generally, it is difficult to create a new framework other than the currently provided indicators, for example, to realize a classification (e.g., classify emotions into four categories: joy, anger, sorrow, and pleasure) using biometric sensor data, due to the number of data channels, resolution, and lack of accuracy. Therefore, it is often easier to obtain results by reducing the evaluation to a one-dimensional evaluation (evaluating whether the degree of enjoyment is strong or weak) by narrowing down the conditions depending on the scene.
The following is a list of assumed difficulty levels for each analysis framework, which is completely subjective.
Framework | Difficulty(1: Difficult…5: Easy) |
---|---|
Classification of Movement | 2 |
Physical Activity Intensity Indexing | 4 |
Classification of mental activity such as emotion | 1 |
Classification of mental activity intensity such as immersion | 3 |