The sleep-stage analysis software prototype developed under this grant is a full-featured, graphically-oriented Microsoft Windows application. ARK uses a combination of Period/Amplitude Analysis, Fast Fourier Transform techniques, and the newer Fast Orthogonal Search algorithm to identify and extract the short "signature" waveforms commonly encountered in sleep EEG data: sleep spindles, K complexes, and saturation artifacts (which simpler approaches often miss). Most existing scoring programs cannot flexibly identify these waveforms, which occupy a substantial part of the Rechtschaffen-Kales criteria for visual scoring, particularly stage 2. ARK reports not only the time of occurrence of those waveforms, but extracts them to graph windows for analyst review. By double-clicking on the sleep-stage diagram (SSD) plot, the analyst can view the region of coupled data sets (EEG, EOG, EMG). from which the waveform was extracted. Thus identified waveforms are visible both in isolation and in situ.

K complexes and sleep spindles are used primarily to determine the occurrence or continued presence of stage 2. False positives (EEG data misidentified as a complex or spindle) don't usually affect the stage 2 analysis as other factors will almost always outweigh the presence of a misidentified signature waveform. False negatives (failure to identify the occurrence of a valid waveform) affect the analysis if too long an interval exists between identified waveforms, as the criteria for "feature dropout"1 will occur. But if ARK finds another K complex or sleep spindle within 3 minutes of the last occurrence, stage 2 will remain the epoch score.

Sleep staging analyses using ARK are usually launched by running a "script" prepared in advance which specifies the names, locations and characteristics of the EEG/EOG/EMG data files. Also specified are the names and locations of "template" data files which contain examples of real or generic K complexes, sleep spindles and (amplifier blocking) artifacts which will be mathematically compared to the data files. ARK finds occurrences of these wave-forms and optionally displays them in a Graph window (WVF) for visual inspection. Each identified wave-form is used in scoring, which is not possible in most existing scoring programs. Identifying these brief "signature" waveforms is essential for scoring stage 2 sleep. ARK also performs more conventional (time domain) Period/Amplitude Analyses and (spectral domain) Fast Fourier Transforms to determine dozens of statistics used in scoring, such as the time in each epoch spent in slow-wave sleep. The graphs of all these measures are displayable versus epoch number (time) in their own Graph Windows, which can be printed, exported to a word processor as bitmap (BMP) files, and manipulated using any of over forty digital signal processing functions, including Steve Pincus's Approximate Entropy. A final graph is produced of the automatically-generated sleep stage diagram (SSD).

The ARK User Manual discusses every menu option and many run-time issues in greater detail, and walks the user through a complete analysis run using supplied tutorial data sets. Over 140 pages long, it will be supplied in print and on media separate from this report.