BCG: mechanical activity measured through the bed AI
A ballistocardiogram (BCG) records small body movements associated with cardiac pumping. In this dataset, two sensor technologies measure mechanical activity through the bed: four electromechanical films beneath the mattress (Film0–Film3) and four load cells beneath the corner bedposts (LC_BCG0–LC_BCG3). Both sets are BCG sensors. Film0 is nearest the head end. Carlson et al., Section 2.1 and Figure 2 show their arrangement.
Which channels enter the report?
The four-feature comparison uses Film0 alone. The 42-feature models use Film0 and LC_BCG0, alongside ECG and PPG. A channel number identifies a sensor; it is not a preprocessing stage. Multiple channels observe different mixtures of the same underlying mechanical activity, with different coupling and signal quality.
In the feature detector, J is a prominent peak 100–400 ms after an ECG fiducial; I and K are neighboring local minima. Film0 and LC_BCG0 are sign-inverted following the reproduction’s polarity convention. These approximate landmarks define timing and voltage differences; they are not independently validated physiological events in every beat.
The source paper evaluated each sensor separately with the same feature definitions. Its term multiple predictors refers to multiple features from one sensor, rather than all sensors jointly. My maintained comparison retains the stated channel subset. See the source-study comparison and waveform features.