Instead of simply excluding contaminated data from the analysis, artifacts can sometimes be removed or significantly suppressed by using methods for signal decomposition such as signal space projection (SSP; Uusitalo and Ilmoniemi, 1997) or independent component analysis (ICA, see Section 3.1 below). The assumption behind the SSP method is that artifacts are confined to a small-dimensional spatial subspace with specific topographic patterns that are orthogonal or almost orthogonal to the brain signal patterns of interest and can thus be suppressed with appropriate projection vectors. Projection vectors can be derived from instances of Raw as well as Epochs. MNE-Python also offers command-line level scripts and Python-level functions to automatically detect heart beats and eye blinks in the data, making automatic SSP computation possible. Once projection vectors are specified for subtraction in the measurement info, MNE minimizes memory and disk space usage by not modifying the original data but instead applying the projections on demand. This enables the user to explore the effects of particular SSPs later in the pipeline and to selectively abandon some projection vectors if the signals of interest
