--- Reviewer B
I find the article very well written and the covered scope almost perfect. I would like to suggest a few extensions and modifications.
.. the origin of the magnetic fields ... It is commonly agreed that the postsynaptic potentials cause a so-called impressed current which may cause magnetic field changes outside of the human head when a sufficiently large number of neighbouring pyramidal cell experiences this a current at about the same time. But, it might be worth to mention that alternative sources of the MEG/EEG are also considered, e.g. action potentials. Action potentials are usually much shorter in time and traveling through space. Additionally, the action potential can be physically described as two opposing current dipoles in close proximity which leads to a stronger magnetic reduction with growing distance than a single current dipole. Therefore, it seems to be impossible to superimpose the fields of a number of action potentials to make them measurable. Recently, the community started to discuss scenarios under which the effect of the action potential is better described by a single spatially stationary dipole instead of a moving pair of dipoles: a) The axon passes a change in the extent of the conducting volume (small diameter to large diameter, e.g. fibres from the spine enter the skull). b) A starting or arriving action potential consists of a single non-moving dipole. In summary, I would suggest to mention that there is a debate about further possible origins of the MEG.
- A flux transformer consists of a sensing and a coupling coil plus wires between the two. It might be good to add the coupling coil and eventually even the SQUID to the figure? The message 'to the SQUID' should be extended into 'to the coupling coil and SQUID' or similar.
- Magnetic field changes are inducing a voltage within each loop of the flux transformer. The voltages of all loops together may cancel out each other or not, but the net voltage is driving the same current in both loops, which may become clearer if the circuit of the flux transformer is displayed completely.
- Modern MEG devices appear to have either magnetometers, radial gradiometers or planar gradiometers or a mixture of the above as sensing coils. Therefore, I would prefer to consider all three types as the most popular. It would mean to mention planar gradiometers. Certain software (e.g. fieldtrip, Donders, Netherlands) offers an option to transform MEG data measured using any sensing coils into data of virtual planar gradiometers to achieve a single peak in the magnetic field strength for a single current dipole (avoiding the two extremes of magnetic field distribution when using magnetometers and radial gradiometers).
- Concerning the achievable balance: The definition of gradiometer imbalance might be included in the article: An imbalance f=1% means that the gradiometer measures 1% of a perfectly homogenous field strength instead of zero. Nurminen, Taulu & Okada (2008) PhysMedBiol discuss this more deeply and state that the typical imbalance is 0.5% when using wires to make the coils (references to further papers are given). Using thin film technology a imbalance smaller than 0.1% is typically achieved.
Software noise cancellation
- Very crucial for these powerful techniques is the ability of the MEG device to precisely measure the interferences. If the dynamic range of the MEG is insufficient the following software cancellation must fail.
- Adding a reference array in a distance to the head provides more power than just compensation of the imbalance. As stated in the paragraph and shown by Vrba it allows to construct third order gradiometers, which provide a much stronger interference suppression compared to the first order gradiometers. It is not necessary that the reference array is completely insensitive to brain signals and solely influenced by the interferences.
- The decomposition by Taulu et al. is based on expanding the measured field distribution into a Laurent series using derivatives of spherical harmonics as basis functions. The series converges within a volume that contains at least all MEG sensors, because this volume is usually free of magnetic sources. The series consists of terms with positive and negative degree. Therefore, one can split the series into two parts and interpret them mathematically as contribution of the inner volume (mainly brain) and the outer volume (mainly interferences). Coefficients of all included basis functions are chosen such that their superposition fits best the real data. A reconstruction of a magnetic field distribution while ignoring the coefficients for the outer basis functions yields a significant interference suppression. In other words, to my understanding it is more the properties of the Laurent series than the properties of the magnetic field distribution which is critical for this method.
- The MEG is sensitive to currents. It is also sensitive to the volume conduction currents within the brain. Otherwise, there would be no need to model a volume conductor. I would rather stress the increase in conductivity when traveling from the brain through the skull. Therefore the strength of the volume currents within the skull becomes significantly smaller and their impact to the localization is small. - EEG measures volume currents - I find this statement incorrect. The EEG measures an electrical potential which is produced by a distribution of electrical charges. Volume currents will alter this charge distribution.
Minimum norm based approaches
- Depth weighting / noise or sensitivity weightings and other norms were suggested to overcome some of the disadvantages. However, I agree, that even this does not solve the problem that our brains most likely do not work at any of these minimums.
The article is clear and concise covering the main issues in MEG. As a general comment, the article could benefit from more visual illustrations. I would also like to suggest the following minor additions:
1- The authors may consider including a paragraph on MEG field maps and their relation to underlying source distributions.
2- The article would also benefit from discussing the relative importance of the time series recorded at the sensors and the influence of noise in source localisation. The inverse problem begins with the recorded time series and not from the scalp distribution of sources, as is some times assumed. Small amounts of sensor noise can have significant effect on source localisation. How is this noise dealt with? A short description of pre-processing steps may also be relevant and useful.
3- The above could lead to a brief description of spatial filter approaches as a means of reducing unwanted noise but also as a solution to the inverse problem. The authors have briefly described dipole fit and minimum norm approaches and a brief description of spatial filter (beamformer) approach would not be out of place here because of their increasing popularity and also given the authors’ own track record and expertise with this approach.
The contents are well written
In statistics for MEG images
They missed SAM synthetic aperture magnetometry
One of authors, Dr. Hirata has several papers about this method
They should add
References are too small numbers, I suggest they should add more references, particularly clinical applications