Common Mistakes and Errors

(and how to avoid them)

© 2009-2011 Kayser

last updated: June 10, 2019

	CSD toolbox home
	CSD toolbox tutorial
	common mistakes and errors
	frequently asked questions FAQ

Errors	May 19, 2009	1. EEG montage does not reflect the EEG/ERP data
	May 22, 2009	2. Generating CSD estimates from CSD data
	June 2, 2009	3. Generating CSD estimates from EEG power spectra
	June 3, 2009	4. Including data of a linked or non-cephalic reference
	February 22, 2011	5. Equating CSD estimates to surface potentials ("CSD reference")

1. EEG montage does not reflect the EEG/ERP data

The most common mistake is to use an incorrect EEG montage, either for the submitted surface potentials or to generate the transformation matrices G and H (i.e., the sequence of scalp locations does not adhere to the surface potential data matrix submitted to the CSD toolbox routines GetGH.m or CSD.m). This error comes in many flavors and may have very different origins. If an incorrect montage E is applied to the ERP data used in the CSD toolbox tutorial (see ExampleCode1.m), for example, by having the Nose site listed first instead of last, this mistake will go unnoticed in the 2-D topography created with MapMontage.m because all 31 sites are correctly represented and mapped. However, the G and H transformation matrices will not match the data, resulting in wrong CSD estimates. Correct EEG Montage Incorrect EEG Montage Top row: Topographies for sample point 21 (100 ms; N1) of the original nose-reference surface potentials (ERP [µV]) and the corresponding surface Laplacians (CSD [µV/m²]) using the correct EEG montage. Bottom row: ERP and CSD topographies of same data when applying an incorrect EEG montage (i.e., channel sequence is out of order). Mapping both ERP and CSD data using the same (incorrect) EEG montage may help to recognize this mistake in time before the Rorschach principle of free association can take over during attempts to interpret the "CSD" topography. In the maps shown above, neither a typical N1 topography for ERPs nor an N1 sink topography for CSDs are present, which should immediately set off alarms and raise red flags, even for ERP researchers unfamiliar with CSD transformations. Inspecting the data and location of the reference, which should show zero values for the ERP data, is always a good face validity check. However, it should be noted that some EEG montage errors may not always be self-evident, although the resulting CSD estimates are nevertheless equally wrong. It is essential for valid applications to be absolutely confident that the correct EEG montage is applied throughout these transformations.

Back to top	Home	Tutorial	Errors	FAQs

2. Generating CSD estimates from CSD data

Another common error is to inadvertantly submit a CSD data matrix to the CSD transform. Whereas the algorithm can easily perform this computation, the transformation itself is meaningless. While an experienced ERP/CSD researcher will likely recognize the suspiciously focal sink-source pattern as an artifact, particularly when using a rigid spline interpolation (i.e., constant m = 4) with λ smoothing, or the usually large CSD values, these artifacts may not be evident to someone with less experience. Again, a good protection against this mistake is to inspect and verify the input data to make sure that only surface potentials are submitted to the CSD toolbox function CSD.m. Topographies for sample point 21 (100 ms; N1) of the original nose-reference surface potentials (ERP [µV]), the corresponding surface Laplacians (CSD [µV/m²]), and "surface Laplacian" estimates derived from the CSD data (CSD-to-CSD). The latter transformation has no physiological meaning.

Back to top	Home	Tutorial	Errors	FAQs

3. Generating CSD estimates from EEG power spectra

A fatal mistake is computing "spectral current source densities" from EEG power spectra, which also renders a meaningless transformation. Because the power spectrum itself is a nonlinear and irreversible data transformation (power is a squared measure, and a power spectrum "rectifies" negative values; see Fig. 1 in Tenke & Kayser, 2005 reproduced below), power spectra can neither be re-referenced (i.e., power spectra computed using a linked-mastoids reference can not be converted to nose-referenced power spectra by subtracting the nose channel from all other channels in the EEG montage) nor can these data be used as input to the CSD toolbox function CSD.m. Even an experienced ERP/CSD researcher will not be able to recognize this error as such from the transformed data (i.e., the "CSD spectra"), although they will likely not display CSD topographies characteristic of alpha or beta frequency bands as described in this article: Tenke, C.E., & Kayser, J. (2005). Reference-free quantification of EEG spectra: combining current source density (CSD) and frequency principal components analysis (fPCA). Clinical Neurophysiology, 116(12), 2826-2846. doi:10.1016/j.clinph.2005.08.007 Why power spectra cannot be rereferenced or used for CSDs [cf. Fig. 1 from Tenke & Kayser, 2005, p. 2829; caption modified for emphasis]: (A) Arbitrary sinusoidal signal (solid lines) with an amplitude topography varying linearly along the midline from zero at FPz to a maximum peak of 3 µV at Pz. Amplitudes of the sinusoid at lateral locations are identical to those at the midline, with the exception of an imposed asymmetry at F3/4 (right greater than left). Rereferencing the sinusoid to Cz reverses the frontal asymmetry (dashed lines). (B) Differences of waveforms shown in A. Because rereferencing is a linear operation, differences between the original and the rereferenced signals are identical at all sites. (C) When the waveforms shown in A are rectified (i.e., converted to absolute values, analogous to power), the transformed waveforms maintain properties of the original sinusoidal signal, including the reversed asymmetry at F3/4. (D) The difference between rectified waveforms changes considerably across the montage, because rectification is not a linear transformation. Likewise, a power transformation (i.e., squared amplitudes) rectifies the signal and is nonlinear. However, the raw EEG epochs can be converted to CSD epochs using the same transformation (i.e., function CSD.m), and valid CSD spectra can be computed from these CSD epochs, for instance, by means of FFT.

Back to top	Home	Tutorial	Errors	FAQs

4. Including data of a linked or non-cephalic reference

Surface potential data stemming from references that do not correspond to physical scalp locations can not be modeled using any single scalp site. For example, if a linked recording reference was used (e.g., linked mastoids, linked ears), this averaged activity cannot be included as a reference channel in the EEG/ERP data matrix that is submitted to the CSD toolbox function CSD.m. In contrast, the data from each of the original reference locations could be included if, and only if, they are both available (as explained here). However, it is perfectly valid to use rereferenced surface potential data without including a reference channel in the montage. This consideration also applies to data rereferenced to the average or common activity of all channels in the EEG montage (i.e., when using an average or common reference), or when using a non-cephalic reference (e.g., sternum).

Back to top	Home	Tutorial	Errors	FAQs

5. Equating CSD estimates to surface potentials ("CSD reference")

Transforming surface potentials [µV] into current source density [µV/cm²] estimates results in a truly reference-independent measure. However, this transformation is not merely a re-referencing strategy, which is always reversible. For example, nose-referenced EEG can be re-referenced to linked mastoids or a common average, and any of these transformations can easily be reversed as long as the original EEG montage is kept intact (i.e., no channels are dropped). In contrast, CSD estimates can not be converted to surface potentials, that is, the CSD transform is irreversible.

Back to top	Home	Tutorial	Errors	FAQs