 
Comparison of a stage diagram from ARK with one derived from visual scoring is made below. Movement Time (MT) is at the top, below it is Waking, then REM and stages 1, 2, 3, 4. In ARK, clicking anywhere on the hypnogram scrolls the relevant EEG data (C4/A1 or C3/A2) with synchronized EMG and EOG traces to the approximate region of interest, which facilitates visual review of the stage decision.
The automated stage diagram generated by ARK retains the overall structure of the visually-scored diagram, with deep sleep primarily in four 40-minute cycles during the session first half, and numerous short waking periods in the second half. The ARK stage diagram also has considerably more structure, perhaps because visual scorers leaf back and forth quite a bit to integrate "outlier" scores, a visual form of averaging. There is considerably more stage 3 slow wave sleep than the visual scoring shows (also see the plot of time in slow wave sleep, tHASW, below) and examination of the EEG (by clicking on the ARK stage diagram, the montage is scrolled to the relevant epoch) reveals that slow wave is present above the thresholds both in duration (20 % of epoch for stage 3, 50% for stage 4) and voltage (75 microvolts peak-to-peak):
This is a visually scored sleep stage diagram from the same data used by ARK. There are only two very brief deep sleep intervals:
Note the Stage REM and Stage 3 (slow-wave sleep) excess over visual scoring noted above:
While examining plots versus epoch of several characteristic voltages, half-power band frequencies f1/2, or event counts, several "heuristics" (purely phenomenological indicators) were uncovered which correlate exceptionally well with known stages but are not mentioned in R&K, and which may have some validity as useful physiologic markers. Often on examining them phrases like "relatively elevated" or "broad local minimum" come to mind. While not incorporated into the decision process in ARK, they probably should be. It would be straightforward to design algorithms to include these indicators from smoothed versions of the epoch values. For example, the graph of fEEG vs epoch (f1/2 is that frequency at which half the band energy is above f1/2, and half below) exhibits spectral slowing on the descending ramps during deep sleep, inevitably followed by rapid speed-up into waking. The fastest elevated regions exist during waking or REM epochs. But even after waking in the middle of deep sleep the slowing continues from nearly where it was interrupted, not from the peak values:
|
