Physiological Modeling 29.6-4 ANALYSIS OF BRAIN SYNCHRONIZATION, BASED ON NOISE-DRIVEN FEEDBACK MODELS Bob Kemp Alpo Varri, Agostinho da Rosa Kim D. Nielsen, John Gade, Thomas Penzel 1. Department of Clinical Neurophysiology, University Hospital Leiden, The Netherlands ABSTRACT The activity of many brain cells can be synchronized by synaptic interconnections. A noise driven feedback model of this mechanism simulates both rhythmic activity and phasic events in the EEG. While classical algorithms such as Fourier analysis quantify power, the model-based algorithms quantify the synchronized, that is predictable, part of this power. The model-based method shows better time resolution and is biased less by artifacts and randomness. MODEL OF BRAIN SYNCHRONIZATION. Many cells in a cerebral network will be synchronized if there are many and/or strong synaptic contacts. The synchronized group activity cycles through the volume of interconnecting pathways, producing rhythmic activity in the EEG. The hardware of these pathways determines the cycle duration, and thus the frequency of the EEG rhythm. In the model [1,2] (figure I), the pathways are represented by a feedback loop (G), through which the synchroni zed fraction (s) of the cell group activity (du(t)/dt) cycles back to itself. In a slow wave model, the feedback would pass mainly 1 Hz activity. The strength of the synaptic coupling in the pathways determines the amplitude of the cycling activity and thus the amplitude of the EEG rhythm. In the model, this strength is represented by the feedback gain (x), usually between 0 and 1. White noise, dw(t)/dt, drives the model. The lowpass filter, L, simulates volume conduction of cerebral activity to scalp EEG. du(t)/dt .1 L Figure 1: Model of brain synchronization. SIMULATIONS Simulations of sigma- and alpha rhythms by such models can hardly be distinguished from real EEG [2,3]. When the feedback gain is 0 (i.e. 0% coupling), a low voltage desynchronized eeg is simulated. Increasing the gain (up to I, i.e. 100% coupling) resul ts in an increasing, but waxing and waning, rhythmic component. The waxing and waning is caused by the noise, which randomly tends to increase or decrease synchronized activity in the feedback loop. Transient EEG changes, i.e. phasic events, can be simulated [4,5,6] by adding pulses to the input noise. This will add an invariable waveform to the EEG. Pulses added to the coupling factor (feedback gain) will result in a component of variable shape, because these pulses are multiplied by the stochastic, synchronized acti vi ty in the feedback loop. This variable response will have about the same main frequency as the synchronized tonic eeg activity. Fixed block pulses administered at random to the slowwave model resulted in variable responses very much like the variable K-complexes seen in real sleep-EEG. Increasing the pulse rate causes the responses to superimpose; the result can hardly be distinguished from real EEG during deep slowwave sleep [5,6]. . _ ...... . ..... ' Fig.2: Simulations of, from top to bottom, alpha rhythm, sleep spindles, K-complexes and slowwavesleep EEG. Input pulses dotted. Annual International Conference of the IEEE Engineering in Medicine and Biology Society. VoL 13. No.5. 1991 CH3068-4/9110000-2305 $01.00 © 1991 IEEE 2305