![Example Java code
• double dbSleep = 0.0033 * (1.06 * ZCM[i - 4] +
• 0.54 * ZCM[i - 3] +
• 0.58 * ZCM[i - 2] +
• 0.76 * ZCM[i - 1] +
• 2.3 * ZCM[i] +
• 0.74 * ZCM[i + 1] +
• 0.67 * ZCM[i + 2]);
• if (dbSleep > 1) {
• sleep[i] = 0;
• } else {
• sleep[i] = 1;
• }](https://image.slidesharecdn.com/0356e026-bac4-47cd-b0d4-fa7ad9fd93f4-160312233708/85/WRAIR-26-320.jpg)
WRAIR
- 1. Development of New Motionlogger Actigraph Martin Bruner, MSEE Bruner Consulting Inc. marty@bruner-consulting.com (720) 494-2545
- 2. Purpose • Describe the science of actigraphy • Describe the Motionlogger Actigraph • Identify Motionlogger Design Issues and Trade offs. • Describe new product development
- 3. About Martin Bruner • President, Bruner Consulting Inc. • Master’s Science Electrical Engineering • Designed Actigraphs and support Software since 1990 • Partners with Ambulatory Monitoring Inc.
- 4. What is Actigraphy? • Actigraphy is a non-invasive method of monitoring human rest/activity cycles. A small actigraph unit, also called an actimetry sensor, is worn for a week or more to measure gross motor activity. The unit is usually, in a wrist-watch-like package, worn on the wrist. The movements the actigraph unit undergoes are continually recorded and some units also measure light exposure. The data can be later read to a computer and analyzed offline; in some brands of sensors the data is transmitted and analyzed in real time.
- 5. Uses for Actigraphy • Sleep Medicine • Rest/Activity Cycles – Shift Work Schedules – Chronotherapeutics(Oncology) • Psychiatry – ADHD – Mood Disorders – Eating Disorders • Fatigue/Performance • Neurology – Movement Disorders – Parkinson’s Disease
- 6. Sleep Medicine/ Detection of Sleep Apnea events using Actigraphs
- 7. Affect Broken Sleep has on Patient Fatigue
- 8. What is an Actigraph (Motionlogger) • A wrist worn device used to measure human activity. • Motionlogger is usually worn on the non dominant wrist • Data is stored on the device to be later transferred to a host PC or mobile device
- 11. Elements of a Motionlogger • Accelerometer to detect motion • Band Pass filter to select movements within a particular band • Memory for storing data recorded • Interface for communicating with a host computer • Microcontroller to control recording, timing, storage, and transmission to host computer
- 12. Accelerometers • For many years, we used a bimorph piezoelectric cantilever beam for our motion sensor. • Extremely sensitive • Temperature and moisture sensitive • Beam broke often • PIM Drift issues were common • Inconsistencies between units • Threshold detection used for ZCM measurement • Length of beam affects response of device
- 13. Ultra Actigraph
- 14. Solid State Accelerometers • Use of “off the shelf” accelerometers • No drift of signal • Uses significantly more power (300 – 600 uA). • Shorter runtimes between battery changes • Not sensitive to moisture or temperature • Presents three analog signals corresponding to X, Y, Z channels. • Requires digitizing the analog signals • Noise floor for software thresholds calculated during a calibration phase
- 16. Digital Accelerometers • Similar to solid state accelerometers • Data is digitized on the part itself • Many devices have FIFO’s which can store multiple samples (useful for automated sampling) • X,Y, and Z channels are sent using a serial protocol (I2C or SPI) • Generally much lower power than analog solid state accelerometers • Can be very susceptible to noise • Most Over the Shelf monitors (Fitbits) use these types.
- 19. Data recording methods • Zero Crossing Mode (ZCM). The number of Zero crossings are counted during a time interval. Standardized for sleep study. • Proportional Integral Mode (PIM). Area under the motion curve. Useful for energy measurements. • Time Above Threshold (TAT). Amount of time signal is above a particular threshold. This method has been largely replaced by PIM
- 20. Example of Sampling Modes
- 21. 12 Minute Cardio Stress Test
- 22. Epochs • The time interval between stored actigraphy data. • Can vary between applications • One minute epochs is generally accepted for sleep detection • Two seconds is used for Periodic Leg Movement detection. • Shorter epochs give greater resolution, but uses more memory.
- 23. Differences between Scientific Actigraphs and Commercial Sleep Monitors • Accuracy. Scientific Actigraphs are calibrated against Polysomnography (gold standard) • Auto detection of in bed. Scientific Actigraphs generally can tell if a patient is in bed. • Details. Scientific Actigraphs can monitor the patients movement history in much greater detail.
- 24. Activity display on a Typical Subject
- 25. Accurate sleep detection and rescoring algorithm • Historically we have used the Cole/Kripke method. • Developed By Roger Cole, Daniel Kripke, and William Gruen (Founder of Ambulatory Monitoring). • Weighted Moving Average filter on Zero Crossing Data • Rescoring is done after initial sleep detection, using a custom AI algorithm with adjustable rules.
- 26. Example Java code • double dbSleep = 0.0033 * (1.06 * ZCM[i - 4] + • 0.54 * ZCM[i - 3] + • 0.58 * ZCM[i - 2] + • 0.76 * ZCM[i - 1] + • 2.3 * ZCM[i] + • 0.74 * ZCM[i + 1] + • 0.67 * ZCM[i + 2]); • if (dbSleep > 1) { • sleep[i] = 0; • } else { • sleep[i] = 1; • }
- 27. How determine the accuracy of an Actigraph? • Polysomnography is considered the “Gold Standard” in evaluating the accuracy of an Actigraph.
- 28. Comparison to Polysomnograpy • Usually performed at a professional or University sleep lab • Patients wear a motionlogger and polysomnography equipment • Sleep data is recorded on both instruments while the patient is “in bed” • Epoch by Epoch comparisons between instruments are performed • A correlation percentage is then calculated.
- 29. Comparing Actigraphy to Polysomnograpy • When comparing data, the sleep results from polysomnography is considered correct. • Actigraphs are judged on their match to polysomnography • Two major elements are important • Sensitivity • Specificity
- 30. Sensitivity • Sensitivity is generally very easy to accomplish. • This parameter measures the actigraph’s sleep determination when polysomnography measured sleep. • Generally most sleep monitors and actigraphs measure in the mid 90%
- 31. Specificity • This parameter measures the Actigraph’s wake detection when polysomnography detects the patient is awake. • This parameter is much more difficult to accomplish. • Most sleep monitors are very poor at this measurement. • Requires a sensitive, low noise accelerometer • High Specificity distinguishes our products from the competition.
- 32. Correlation to Polysomnography • Historically, our motionlogger actigraphs have correlations between 87% and 92% correlation with polysomnography. • Most of our competitors are below 85% • Fitbits usually run below 70% • Generally, the more restless the sleep, the lower the correlation to polysomnography.
- 33. Fitbit comparison to Motionlogger
- 34. Fitbit comparison to Motionlogger • Fitbit grossly over predicted sleep • Awake algorithm over compensates the awakings. • Fitbit missed the start of sleep by 42 minutes
- 35. Elements of a good Actigraph • Low noise environment • Sensitive accelerometer • Precise filtering • Good mechanical energy transfer from wrist to accelerometer • Appropriate sampling rate • Accurate sleep detection and re-scoring algorithm
- 36. Low Noise • Noise is the enemy of accurate actigraphs • The ambient noise represents the minimum threshold for motion detection • The higher the noise, the lower the actigraph’s sensitivity • High current operations can create significant noise in poorly designed circuits
- 37. Noise Example of Actigraphy/Pulse Oximeter Actigraphy Only
- 38. Noise Example of Actigraphy/Pulse Oximeter Actigraphy and Pulse Oximeter
- 39. Significant Noise from Pulse Oximeter • High current required to drive LED • Poor circuit design led to noise being introduced into digital accelerometer • This created the noisy accelerometer output
- 40. Applications to New Actigraph Designs • Wireless transfers can draw fairly high currents while transmitting • If not designed into the system, this current drain can significantly affect the noise level in the accelerometer • Fitbit has significant noise due to periodic BLE advertisements (two seconds).
- 41. Actigraphy Noise Considerations • Avoid taking data during high current operations. • Pay careful attention to printed circuit board design • Ensure that the accelerometer glue components are correct. • Use a very stable power supply
- 42. Avoid taking data during high current operations • If at all possible, keep higher current operations to a minimum. • In previous designs, we shut off data collection during data transfers. • This technique generally requires the user to initiate data transfers (button press).
- 43. Pay careful attention to printed circuit board design • Careful routing of ground, Vdd, Vdd_IO, and AVdd • Prefer separate planes for Vdd and Vss • Separate regulators for Vdd, AVdd, and Vdd_IO would be preferable (maybe not practical)
- 44. Ensure that the accelerometer glue components are correct
- 45. Use a very stable power supply • Low Noise DC-DC converter very important. • LDO (low drop off) regulators with high Power Supply Ripple Rejection (PSRR) work well. • Often switching regulators will introduce additional noise into that system. • However, the low efficiencies of LDO’s may require the use of a switching DC-DC converter.
- 46. Sensitive Accelerometer Two factors control the ability of the accelerometer to detect motion: • Acceleration noise density • Sensitivity • Both items determine the minimum threshold that can be detected.
- 47. Precise Filtering • In previous designs, the data from the accelerometer was an analog signal • This analog signal could be amplified and filtered (usually Band Pass 2-3Hz) • Digital filters present a new challenge since the data is digital • High order digital filters (6th order) will need to be used. • Sample rate must be high enough to keep the filters stable.
- 49. Appropriate Sample Rate • For sleep detection, human motion between 2-3 Hz is used. • This means 6 Hz sample rate minimum (Nyquist Criterion). • However, we generally use higher frequencies for off wrist detection, Periodic Leg Movements, and Seizure detections. • Last product had a sample rate of 32 Hz. • Would like to try an ODR of 50 Hz in the future. This will permit the use of higher order digital filters.
- 50. Current Products • AMI Currently Markets three Actigraphy devices • Motionlogger Actigraph • Micro Motionlogger Actigraph • ZZZ-Logger • All use IrDA for data transfers
- 51. Motionlogger
- 52. Motionlogger Features • ZCM, PIM, Light, Temperature • PVT, Mood Scale, Timer, Alarm • 30 Day Run Time • Removable Battery
- 53. Motionlogger Issues • Bulky design, uncomfortable to wear • Uses non standard 2450 battery • Button Breaks • Screw inserts come loose • Poor water retention • Buttons have tendency to fail • Designed Micro to address issues
- 55. Micro Motionlogger • Uses similar electrical design to Motionlogger • 30 day run time • Accelerometer is lower powered an more sensitive, meaning less signal conditioning components • ZCM, PIM, Light, Temperature, Events, Life Measures • Uses 2430 battery • Membrane switches improves water retention and switch durability.
- 56. ZZZ-Logger
- 57. ZZZ-Logger • Hardware identical to Micro • Different firmware to support mobile devices (currently only Android) • Sold in Japan as Sleep Watchman