Actigraphy-analysis-project - Medicoimpianti

Vai ai contenuti

Menu principale:

Actigraphy-analysis-project

    
Actigraphy analysis project
Predictive algorithms of individual wrist actigraphy for real life.


The project.
This project is looking for new ways of sleep/circadian data analysis.  
The main target is to search for predictive algorithms from wrist actigraphy recordings of one subject using a MotionWatch8 (CamNtech Ltd) system on the non predominant wrist, for unrestricted lifestyle.  

The context
Wrist actigraphy has been used in the past 30 years to monitor motion activity.  
The actigraphy data have been accepted for the analysis of sleep of a single night, with a quantification of sleep duration and its fragmentation.
They are also used to check the presence of fluctuations in the circadian cycle, especially to highlight pathologies with shifts of falling asleep timing.
During the day it is possible to quantify exercise and (with calibration) recognize type and intensity of the exercise.

Continuous monitoring of wrist actigraphy in real life is possible and it is a valuable tool.
Nowadays, the needed hardware is cheap and it is possible to imagine useful applications in real life, for continuous monitoring and smart homes.
There are dozens of startups with wearable monitors based on actigraphy that use more or less the same parameters for recording and analysis of one minute epochs. And then, they make the comparison against a(nother) so called "normal" group they built for the occasion, frozen in time and space.

The goal
Existing analysis methodologies for wrist actigraphy, describe the recorded data but do not provide any personal parameter, I mean something that alert or reassures about your body status, at least for the coming day, as well as we do for Temperature or Blood pressure measurements.
On the contrary, I think that for personal monitors we need personal parameters, that "grow old" with the subject, in his own way.
I think that for such parameters, we need to create data models and develop algorithms that use Artificial Intelligence.
Hopefully, the analysis of the long term, high sampling data of  subproject 1 will allow researchers to find at least one parameter that will be possible to use. Anyway, even if that search fails, the need of the one second epoch in this field of studies is bubbling, and that recording will be useful.
The possible analysis will be tied to the lenght of the recording, but also on the models that will be invented for this new type of recording. It is something that need more than one research center and therefore the idea is to make the first year at one second epoch (Phase 2 and 3) available to the public. Most probably Physionet,  https://physionet.org/  where other actigraphy data are already available.

Sub-project 1 – Dataset

New models and algorithms need long term data sets and guidelines on methodological issues. It seems that both are not easily available, on and off line.  

Some examples are available in partially controlled environments:
- Nearly 600 days in 2 patients
Werth, Esther, Egemen Savaskan, Vera Knoblauch, Paola Fontana Gasio, Eus J.W. van Someren, Christoph Hock, Anna Wirz-Justice. Decline in long-term circadian rest-activity cycle organization in a patient with dementia. J Geriatr Psychiatry Neurol, 2002; Vol. 15; pp. 55-59.
- 50 days and then 30 days after 5 months
Miller, Nita Lewis, Shattuck, Lawrence, G. Sleep Patterns of Young Men and Women Enrolled at the United States military Academy: Results from Year 1 of a 4-Year Longitudinal Study. Sleep, 2005; Vol. 28; No. 7; p. 837.
- One month twice a year for 4 years
Longitudinal Study of Sleep Patterns of United States Military Academy Cadets
Nita Lewis Miller, Lawrence G. Shattuck, Panagiotis Matsangas
Sleep. 2010 December 1; 33(12): 1623–1631.  
- 6 months sea duty
Nita Lewis Shattuck ; Panagiotis Matsangas
A 6-Month Assessment of Sleep During Naval Deployment: A Case Study of a Commanding Officer. Aerospace medicine and human performance Vol. 86, No. 5 May 2015

Few recordings are available in free life:
- 5 months, one patient
Garbazza C, Bromundt V, Eckert A,Brunner DP, Meier F, Hackethal S and Cajochen C (2016)  
Non-24-Hour Sleep-Wake Disorder Revisited – A Case Study.
Front. Neurol. 7:17. doi: 10.3389/fneur.2016.00017
- 4 months, 80 OSA patients and 50 controls.
Sleep remains disturbed in patients with obstructive sleep apnea treated with positive airway pressure: a three-month cohort study using continuous actigraphy  
Sleep Medicine, Volume 24, August 2016, Pages 24-31 Jon Tippin, Nazan Aksan, Jeffrey Dawson, Steven W. Anderson, Matthew Rizzo
- And the amazing 30 years!
ESRS 2016 Bologna
P041 Three decades of continuous motor activity recording:analysis of sleep duration
A. Borbely, T. Rusterholz and P. Achermann Institute of Pharmacology and Toxicology, University of Zurich, Zurich, Switzerland
Objectives: Motor activity recording by a wrist-worn device is a common unobtrusive method to monitor the rest-activity cycle. We present a first analysis of data that have been obtained over more than three decades.

To answer to the need of a suitable data set, I started one long term recording on myself, late 2015. Since it is expected from physiology to find rhythms with a period long at least one year, a recording with a minimum length of two years is today the first target.

Sub-project 1 – Dataset – Phase I
That was the baseline. Six months at 1 minute epoch.

Sub-project 1 – Dataset – Phase II
Six months at 1 second epoch.

Sub-project 1 – Dataset – Phase III
Six months at 1 second epoch.Environnement change: ligh only wake up clock stopped. It was set at 7.00, at about 3 meters from the head.

Sub-project 1 – Dataset – Phase IV
Six months at 1 second epoch. In case of  early morning (4 to 6 a.m) wake up, most of the time  an eye mask is then used.


At the moment it is not known what should be a "sufficient" lenght of such a recording. Since it is the first set ever recorded,  I think it should go ahead as long as possible, because it will show in one subject the change over time that is expected from short recordings in groups of different ages described in literature.  

Subproject 1 - Recording Log
The epoch recording value was set at one minute. For the meaning of data sampling and units see this page.
Beside movement data, there are light levels read by the instrument. Please note that it is not possible to know if and when the instrument is covered by garments.
The Marker option is used to signal when the instrument is taken off. It is inserted at the beginning and at the end of a flat period, sometimes only at the beginning or at the end. It is possible that sometimes the marker button was not pressed properly. If so, it should be only in the morning  or at data download, always only for few minutes.

Phase 2 - Start day 17/06/2016
From now on the epoch will be one second. That will require a data dowload each day and the risk of data loss is higher.
Month 7 - Closed with only 2 days lost.
Month 8 - Closed
Month 9 - Closed
Month 10 - Closed
Month 11 - Closed - Back to standard time-
Month 12 - End

Phase 3 - Start day 17/12/2016
Month 13 - Closed
Month 14 - Closed
Month 15 – Closed
Month 16 - Closed - Change of local time, +1 from the first 3 months
Month 17 - Closed
Month 18 - Closed

Phase 4 - Start day 19/06/2017
Month 19 - Closed
Month20 - Ongoing

The main risk is that the subject will not accept to carry on the recording, for unknown reason. The fact that I'm the researcher and the subject lowers the risk. There are few ways for the subject to influence the data and nearly no one for the more interesting part, that is the sleep at night. That lowers possible bias from the researcher. After the first year, it seems that 10% of raw data are lost due to mismanagement of the recording unit by the subject. The two main reasons are data not downloaded and battery life shorter than expected. For the second, now the battery is changed when it reaches 30% of the full charge.


Additional information
For those not already interested in actigraphy, there is an introduction to the reasons why I find the subject interesting. Could be suitable for an introductory speech for students.
There is a .ppt file Download and related .doc Download

I divid the issue in 3 main areas:
a) Automatics
How to model the situation in which a sensor is located at the far end of a jointed limb of a body in motion with a) limb movement b) movement of the body isolated c) movement of the body from interaction not known d) with position sensor not constant.
Out of past experiences, wrist recording was useful for sleep analysis. In the model we may add, for that situation, the interaction with the mattress and/or the horizontal position.
That could bring modelling of the expected effect, on the recordings, of different sleeping positions.

b) Mathematics
Probably it does exists a specific area, but I would not know which one.
Topic: Transition from a curve in space-time to a discrete one-dimensional numerical series
There are important practical consequences.
While movement recording is more and more widespread due to technological advances, each manufacturer uses his own raw data transformation. Most of them do not declare it, not to mention to document, it.
For that reason, it is impossible to use numerical data out of different systems.
Therefore, each manufacturer of medical devices is providing his own software analysis of the data recorded. Since all of the resulting parameters are somehow validated against gold standard (still visual “readings” of polisomnographic recordings....), the results for that purpose are similar.
A more analytical way to compare raw data could allow better meta analysis of clinical studies that use different devices.

c) AI.  
Out of past experiences, we know that there is a correlation between body movements and body internal states. Which dynamics of internal states is theoretically possible to monitor and which are not, is an always open issue.
For the analysis, the data set reminds me situations close to ethology (recording parameters in a situation not controlled) and astronomy (discrete log of phenomena over long time scales), but of which I have no experience of data processing. I hope that to offer data to the public may generate some contamination.
In .ppt, I left out the part relating to the detection of light because, at this level, I think that conceptually it does not add anything.
First year report - Phase I & II  
While dataset develops, some rough evaluations are possible.  Some are collected in this report. January 2017.
Project PhaseIII report  
Some details in this report August 2017



 
Copyright 2016. All rights reserved.
Torna ai contenuti | Torna al menu