In response to reviewer feedback it's become clear that there is strong interest in acceleration metrics beyond the basic ENMO metric we currently report. One such metric of interest is Monitor-Independent Movement Summary (MIMS). MIMS units provide an open-source, universal metric for summarizing acceleration data across different devices, eliminating inconsistencies caused by device-specific factors. This standardization enables better cross-study comparisons and supports the integration of both research and consumer-grade devices. Furthermore MIMS has been shown to capture information just above the devices noise threshold, making it more sensitive to subtle movements (as compared to other metrics like Activity Counts).
Wristpy will add a function to the metrics module that will calculate MIMS. Adding MIMS to the Wristpy orchestrator module, and data output will be completed in a future pull request.
Natural cubic spline interpolator to resample to 100 Hz.
[ ] Extrapolation
Find "Maxed- out" samples
Use mathematical model depicted below to find the probabilty that a sample at n is "maxed-out"
Pn (E|x = |d| − d0) = F( x = |d| − d0; k, θ) = γ(k, θx) / Γ(k) if x < 0, Pn (E|x) = 0
Find "Maxed-out" edges
The maxed out regions are determined from the difference between confidence probabilities of adjacent samples. (Q)
Q(n − 1, n) = Pn − Pn−1
If Q is greater than some threshold C (set at 0.5), that sample is a boundary. The sign determines if it is a hill (+) or a valley (-).
Samples within T seconds outside of the left and right boundary are selected for the signal reconstruction
Extend "maxed-out" signal to make it smoother (as though it had not hit the limits of the devices range)
Values within T seconds are modeled using weighted spline regression, where weights are set as probability of not being maxed out. The peak is estimated by finding where the two regression curves meet, and the maxed-out data points are replaced with smoothly fitted values to keep the sampling intervals even.
Description
In response to reviewer feedback it's become clear that there is strong interest in acceleration metrics beyond the basic ENMO metric we currently report. One such metric of interest is Monitor-Independent Movement Summary (MIMS). MIMS units provide an open-source, universal metric for summarizing acceleration data across different devices, eliminating inconsistencies caused by device-specific factors. This standardization enables better cross-study comparisons and supports the integration of both research and consumer-grade devices. Furthermore MIMS has been shown to capture information just above the devices noise threshold, making it more sensitive to subtle movements (as compared to other metrics like Activity Counts).
Wristpy will add a function to the metrics module that will calculate MIMS. Adding MIMS to the Wristpy orchestrator module, and data output will be completed in a future pull request.
Source for algorithm implementation:
Tasks
[ ] Interpolation
[ ] Extrapolation
[ ] Bandpass Filter
[ ] Aggregation
[ ] MIMS Unit Truncation
Freeform Notes
No response