search

xiangnandang / ggir_workshop

public repo for ggir workshop
GNU Lesser General Public License v3.0
1 stars 0 forks source link

Question for Vincent #3

Open juji-lilly opened 4 years ago

juji-lilly commented 4 years ago

Vincent, please see the list of questions from us as follows. Thanks a lot for your help!

  1. If we want to implement another imputation procedure while running GGIR, how can we do that?
  2. What is the difference between UK Biobank and GGIR from your perspective?
  3. In Van Hees et al 2018, the z-angle is a feature used to detect sleep period time (SPT) window, and the raw signal data from each of the axes were used without prior gravity removal. Would that affect the results or it does not matter for detecting the sustained inactivity bouts using the signal data with or without prior gravity removal?
  4. Are the features roll, pitch, and yaw (Ellis 2016) calculated from the three signal axes before applying a low-pass filter to separate the gravity out? But the roll, pitch and yaw angles of the gravity direction were calculated after extracting the gravity from the raw signal on each axis? If so, why not calculating the roll, pitch and yaw directly from the axes after the gravity have been removed?
  5. In part 2 of GGIR, the unreliable signal sections were identified and replaced by imputed values, are the "unreliable signal sections" including only the non-wear sections and clipping sections detected by the algorithm? If not, what else do they include?
vincentvanhees commented 4 years ago

  1. Will discuss as part of the workshop.

  2. UK Biobank is a biobank while GGIR is software. Or do you mean the repository https://github.com/activityMonitoring/biobankAccelerometerAnalysis This software has been developed for processing the UK Biobank accelerometer data. The original version is mostly a translation of GGIR into Python/Java, and was not optimized for processing data from other sensor brands, other timezones or involving different experimental protocols. You would have to checkout out the repo's changelog to know what the latest updates are.

  3. The algorithm looks for change in angle rather than whether magnitude of acceleration has past some threshold, this is it's strength and novelty.

  4. GGIR uses a rolling median filter before deriving the angle as described in the paper. I am not sure I understand the rest of your question. Can you add a link to the Ellis 2016 paper?.

  5. Yes, only clipping and non-wear time.

juji-lilly commented 4 years ago

Thank you @vincentvanhees for your answers. Here is the link to Ellis 2016 paper: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4833514

We also have some other questions for our next workshop on 4/27:

  1. We would like to learn more about embedding external functions: https://github.com/wadpac/GGIR/tree/GGIR2.0/pdf_vignettes_for_github_only
  2. Please explain more about the outputs of the GGIR (e.g. what does "bout" mean? How was it computed and how to use it? And how does it relate to the "sustained inactivity bout"?)
  3. What has been done to validate GGIR? Any data simulation has been conducted? And what area can be enhanced from a statistical point of view? (e.g. imputation method)
  4. Does the algorithm in GGIR part 3 and part 4 follow exactly as the heuristic algorithm to detect the SPT described in Van Hees 2015 (with sleep log) and Van Hees 2018 (without sleep log)?
  5. Is the threshold per individual used for SPT window unique for each day as well? Or we just have one threshold for one individual all the days?
albert-man commented 4 years ago

Hi Vincent,

Some additional questions from the team:

1- What has been done to validate GGIR? Did you conduct any data simulations? 2- What do regulatory agencies in Europe such as EMA think about GGIR tool? 3- Any ideas on how to improve on GGIR computational time? 4- Are you considering adding other features such as Step counts? 5- Is there any support for Empatica Embrace 2 device? 6- Is there any support for other wearable metrics such as EDA?

Thanks and look forward to the workshop today!

Best, Albert

jwang-lilly commented 4 years ago

Albert, the answer to your question 4) is here: https://github.com/wadpac/GGIR/issues/236

vincentvanhees commented 4 years ago

In the workshop today I will present 14 slides in which I respond to the above questions and the request for an elaboration on the sleep detection.

After that we can have an open discussion.