The effect of environmental noise on magnetometer-and gradiometer-based MEG systems

The ability of an MEG system to resolve small or deep sources is solely determined by the achieved signal-to-noise ratio (SNR). Ideal MEG sensors should exhibit a large gain relative to the brain signals and should also effectively attenuate the environmental and brain noise. Designing the system for large signal strength without regard for noise attenuation, or designing for excellent noise attenuation without paying attention to the signal strength does not produce the best results. The simultaneous satisfaction of large signal strength and good noise attenuation is not feasible in practice and the best sensor system represents a compromise that maximizes the SNR. The paper will examine optimal and sub-optimal sensor designs and will illustrate their behavior using a synthetic signal embedded in the measured environmental and white noise. The discussion will be focused on MEG systems based on radial magnetometers or gradiometers, and noise cancellation using references, as in Fig.1.a [1]. In such systems, the primary sensors are located close to the scalp surface and are exposed to the brain signal and the environmental noise. The references are positioned farther from the scalp surface and detect mostly environmental noise. The primary sensors can be radial gradiometers or magnetometers and are both associated with a “baseline”. The baseline for the magnetometers enters the picture because even in conventional shielded rooms, the environmental noise is too large and the magnetometers must be operated with some noise cancellation method [2]. The magnetometer baseline is then a distance between the magnetometer primary sensor and the references and is usually long and varies widely over the scalp surface because the helmet dimensions are large. The primary gradiometer baselines are typically short and are identical for all sensors. For both types of primary sensors, the references are used to either synthesize higherorder gradiometers or adaptively subtract the noise. The optimum primary baseline length is determined by the interplay between the magnitudes of the detected brain signal and the noise [3]. The non-biological noise acting on an MEG system consists of white instrumental and low frequency environmental components, Fig.1.b. The important parameters are the white noise level, νw, the onset of low frequency noise, fob, and the low frequency noise log-log slope, k, defined via νlow = A/f, where A is a constant. The constant A is proportional to the baseline length, A = Aob/bo, where A and Ao correspond to the baselines b and bo, respectively. Thus if the gradiometer baseline is long, the magnitude of the detected environmental noise will be large and the fob will also be large. The contribution of the low frequency noise depends on the lowest frequency of interest, f1, and the bandwidth of the measurement, ∆f, as in Fig.1.b. If f1 is small, low frequency noise plays an important role, while if it is large, the low frequency noise may be negligible. For a constant f1, the magnitude of the detected environmental noise increases with increasing baseline b (or increasing fob) [3]. Similarly, the detected brain signal magnitude also increases with increasing baseline [3]. Since the functional dependencies on the baseline of the detected noise and the brain signal are different, the SNR (which is their ratio) exhibits a maximum as a function of the baseline. The SNR maximum defines the optimum operating point for the MEG system, Fig.2.a. The optimum baselines are typically short, in the range from about 2 to 8 cm and they depend on the environmental noise at a particular site [3]. Typical noise parameters encountered within shiel-