Time-Domain Analysis

Traditional display of the electroencephalogram is a graph of biopotential voltage ( y -axis) as a function of time and consequently is described as a time-domain process. The objective of a diagnostic electroencephalogram is to identify the most likely cause of a detected abnormality at one moment in time. Typically, a diagnostic electroencephalogram is obtained under controlled conditions, using precisely defined protocols. Recorded EEG appearance is visually compared with reference patterns. Interpretation is based on recognition of unique waveform patterns that are pathognomonic for specific clinical conditions. In contrast, the goal of EEG monitoring is to identify clinically important change from an individualized baseline. Unlike diagnostic EEG interpretation, monitoring requires immediate assessment of continuously fluctuating signals in an electronically hostile, complex, and poorly controlled recording environment. Therefore, of necessity, interpretation relies less on pattern recognition and more on statistical characterization of change. Numeric descriptors thus may appropriately form an integral part of EEG monitoring.

Both EEG diagnostic and monitoring interpretations are based in part on the “law of the electroencephalogram” ( Box 12.2 ). It states that amplitude and dominant frequency are inversely related. The inverse relationship between amplitude and frequency generally is maintained during unchanging cerebral metabolic states. Parallel increases in both may occur in some hypermetabolic states such as seizure activity, whereas decreases may be seen in hypometabolic states such as hypothermia. In the absence of these influences, simultaneous decreases in both amplitude and frequency may indicate ischemia or anoxia ( Fig. 12.4 ); a parallel increase may represent artifact ( Fig. 12.5 ).

Box 12.2

Law of the Electroencephalogram

• 
  

  In the absence of disease, electroencephalographic amplitude and frequency are inversely related

  
  
• 
  

  Simultaneous decrease may indicate ischemia, anoxia, or excessive hypnosis

  
  
• 
  

  Simultaneous increase may indicate seizure or artifact

The importance of electroencephalographic (EEG) baseline recording. This two-channel EEG recording was made immediately after induction of anesthesia, before head repositioning for insertion of a central venous catheter. Anesthetic induction apparently uncovered a preexisting asymmetry that was not evident in the waking electroencephalogram. Although the patient had a history of an earlier mild cerebrovascular accident and transient ischemic attacks, he appeared neurologically normal at preoperative assessment.

(Courtesy GE Healthcare.)

Electroencephalographic (EEG) contamination by electrical artifact. The large-amplitude 2-Hz triangular waves in the left frontotemporal derivation (top trace) are the result of temporalis muscle activation with a nerve stimulator. Current spread from the stimulating to the EEG recording electrodes may be minimized with use of the appropriate facial nerve stimulation site at the jaw angle.

(Courtesy GE Healthcare.)

Time-domain analysis of traditional electroencephalography uses linear signal amplitude (ie, voltage) and time scales. The amplitude range of EEG signals is quite large (several hundred microvolts), and univariate statistical measures of its central tendency and dispersion may contain clinically useful information. Furthermore, amplitude variation may show clinically significant changes in reactivity that can be obscured by frequency-domain analysis. Advances in the technology of EEG amplitude integration have prompted a resurgent interest in this attractively simple approach, particularly in pediatrics.

Frequency-Domain Analysis

An alternative method, frequency-domain analysis, is exemplified by the prismatic decomposition of white light into its component frequencies (ie, color spectrum). As the basis of spectral analysis, the Fourier theorem states that a periodic function can be represented in part by a sinusoid at the fundamental frequency and an infinite series of integer multiples (ie, harmonics). The Fourier function at a specific frequency equals the amplitude and phase angle of the associated sinusoid. Graphs of amplitude and phase angle as functions of frequency are called Fourier spectra (ie, spectral analysis). The EEG amplitude spectral scale ( Fig. 12.6 ) squares voltage values to eliminate troublesome negative values. Squaring changes the unit of amplitude measure from microvolts to either picowatts (pW) or nanowatts (nW). However, a power amplitude scale tends to overemphasize large-amplitude changes. Clinically important changes in low-amplitude components that are readily discernible in the linearly scaled unprocessed EEG waveform may become invisible in power spectral displays.

Comparison of time- and frequency-domain electroencephalographic (EEG) displays. The traditional analog EEG signal shown in the upper left is a time-domain graph of scalp-recorded amplitude (µV) as a function of time. Digitized EEG segments (epochs) are computer-processed using the fast Fourier transform (FFT), which, like a prism, decomposes a complex electromagnetic signal into a series of sinusoids, each with a discrete frequency. The instantaneous relationship is then graphically depicted by the power spectrum (lower left), a frequency–domain plot of power ( µV ² or pW ) as a function of frequency. The spectral edge frequency (SEF) defines the signal amplitude upper boundary. The three-dimensional compressed spectral array (CSA) plots successive power spectra with time on the z -axis (upper middle). The density-modulated spectral array (DSA; upper right ) improves data compression by using dot density to represent signal amplitude (ie, power). Amplitude resolution is improved through color coding in the color density spectral array (CDSA) shown at the lower right. The SEF is shown as the white vertical line. Note the EEG suppression at the bottom of each spectral trend.

Simplification of the large amount of spectral information generally has been achieved through the use of univariate numeric descriptors. Most commonly, the power contained in a specified traditional EEG frequency band (delta, theta, alpha, or beta) is calculated in absolute, relative, or normalized terms.

The most widely used univariate frequency descriptors ( Box 12.3 ) are as follows:

• 1.
  

  Total power (TP).

  
  
• 2.
  

  Peak power frequency (PPF), the single frequency of the spectrum that contains the highest amplitude.

  
  
• 3.
  

  Mean dominant frequency (MDF), the sum of power contained at each frequency of the spectrum times its frequency divided by the TP.

  
  
• 4.
  

  Spectral edge frequency (SEF), the frequency below which a predetermined fraction, usually 90% or 95%, of the spectral power occurs.

  
  
• 5.
  

  Suppression ratio (SR), the percentage of flat-line electroencephalogram contained within sampled epochs.

Box 12.3

Common Univariate Electroencephalographic Descriptors Detecting Ischemia

• 
  

  Total power (TP)

  
  
• 
  

  Peak power frequency (PPF)

  
  
• 
  

  Mean dominant frequency (MDF)

  
  
• 
  

  95% spectral edge frequency (SEF)

  
  
• 
  

  Suppression ratio (SR)

Scalp-recorded cerebral biopotentials are complex physiologic signals. They represent the algebraic summation of voltage changes produced from cortical synaptic activity (ie, electroencephalogram), upper facial muscle activity (ie, facial electromyogram [fEMG]), and eye movement (ie, electro-oculogram [EOG]). During consciousness and light sedation, high-frequency gamma power (ie, 25–55 Hz) is a mixture of electroencephalography and subcortically influenced facial electromyography. Muscle activity makes a larger contribution because of the closer proximity of signal generators to the recording electrodes. Hypnotic and analgesic agents typically suppress both cerebral and muscle activities, with resulting reduced gamma power. Because the upper facial muscles are relatively insensitive to moderate neuromuscular blockade, they may remain reactive to noxious stimuli. Nociception results in sudden gamma power increase, independent of activity in the lower-frequency classic EEG bands.

The EEG analyzers just described either provide separate quantitative estimates of the high-frequency information or incorporate this information into the hypnotic index. For example, the Datex-Ohmeda Entropy Module (GE Healthcare/Datex-Ohmeda, Helsinki, Finland) separately analyzes the 32- to 47-Hz band and terms the signal Response Entropy (RE). Addition of RE to the lower-frequency state entropy (SE) is claimed by the manufacturer to facilitate distinction between changes in hypnosis and analgesia, although supporting evidence for this proposition awaits carefully designed and adequately powered randomized, prospective studies. EEG suppression decreases both entropy indices because noise-free flat-line EEG segments are generally thought to have near-zero entropy. However, during cardiac surgical procedures, EEG signals that appear to be totally suppressed may be associated with paradoxically very high entropy values. To minimize this problem, SE uses a special algorithm that assigns zero entropy to totally suppressed EEG epochs.

In addition to the quantitative EEG numeric indices, many monitors also display pseudo-three-dimensional plots of successive power spectra as a function of time. This frequency-domain approach was originated by Joy and was popularized by Bickford, who coined the term “compressed spectral array” (CSA). Its popularity stems in part from enormous data compression. For example, the essential information contained in a 4-hour traditional EEG recording consuming more than 1000 pages of unprocessed waveforms can be displayed in CSA format on a single page.

With CSA (see Fig. 12.6 ), successive power spectra of brief (2- to 60-second) EEG epochs are displayed as smoothed histograms of amplitude as a function of frequency. Spectral compression is achieved by partially overlaying successive spectra, with time represented on the z -axis. Hidden-line suppression improves clarity by avoiding overlap of successive traces. Although the display is aesthetically attractive, it has limitations. The extent of data loss resulting from spectral overlapping depends on the nonstandard axial rotation that varies among EEG monitors.

An alternative to the CSA display to reduce data loss is the diversity-modulated spectral array (DSA) that uses a two-dimensional monochrome dot matrix plot of time as a function of frequency (see Fig. 12.6 ). The density of dots indicates the amplitude at a particular time-frequency intersection (eg, an intense large spot indicates high amplitude). Clinically significant shifts in frequency may be detected earlier and more easily than with CSA. However, the resolution of amplitude changes is reduced. Therefore color DSA (CDSA) was developed to enhance amplitude resolution (see Fig. 12.6 ). The CSA, DSA, and CDSA displays are not well suited for the detection of nonstationary or transient phenomena such as burst suppression or epileptiform activity.

In summary, a quick assessment of EEG change in either the time- or frequency-domain focuses on (1) maximal peak-to-peak amplitude, (2) relation of maximal amplitude to dominant frequency, (3) amplitude and frequency variability, and (4) new or growing asymmetry between homotopic (ie, same position on each cerebral hemisphere) EEG derivations. These objectives are generally best achieved through the viewing of both unprocessed and processed displays with a clear understanding of the characteristics and limitations of each ( Box 12.4 ).

Box 12.4

Measures That Define Electroencephalographic Changes

• 
  

  Maximum peak-to-peak amplitude (or total power)

  
  
• 
  

  Relation of maximum amplitude to dominant frequency

  
  
• 
  

  Amplitude and frequency variability

  
  
• 
  

  Right-to-left symmetry