Example of global artifacts appearing in all the recording channels at same temporal window. The data are recorded from the hippocampus of a rat. 

Context in source publication

Context 1

... this paper, first of all, the characteristics of labeled artifacts appeared in in vivo neural recordings are analyzed and accord- ingly those artifacts are classified into four types. Subsequently, an artifact detection and removal algorithm is proposed. The algorithm relies on the spectrum characteristics of the neural signals (i.e. LFP and neural spikes) for artifact detection. It further applies stationary wavelet transform (SWT) to detect possible artifactual regions from the decomposed wavelet coefficients. Once artifacts have been detected, to restore neural signals, a modified version of the existing universal-threshold value is proposed, which makes the algorithm more robust. In order to validate the proposed algorithm with quantitative measures, extensive simulations have been performed on both real and synthesized data. The rest of this paper is organized as follows. Section 2 gives the problem description. Section 3 focuses on formulation and analysis. In Section 4 comparative simulation results are presented. Section 5 provides discussions about the performance of the proposed algorithm. Section 6 gives concluding remarks. In this section, the characterization of different artifacts, their sources and properties are presented. Such characterization efforts help to develop a better algorithm and generate synthesized neural database for performance assessment of the algorithm. Artifacts in a broad sense can be classified into two categories: local and global . Local artifacts are localized in space, i.e. appear only in a single recording channel while global artifacts can be seen across multiple channels of an electrode array. An example to illustrate global artifacts is shown in Fig. 1. The artifacts can also be classified into external and internal categories based on their origins (Savelainen, 2010, 2011). Internal artifacts arise from body activities, e.g. due to movements made by the subject itself, sudden changes of bioelectrical potentials, muscle activities, sudden chemical releases, etc. While external artifacts result from cou- plings with unwanted external interferences e.g. power line noise, sound/optical interferences, EM-interferences, etc. The artifacts may appear only once in the whole recording sequence, sometimes they can also appear in a regular/periodic manner. Examples of such artifacts are shown in Fig. 2. There are a number of factors and sources for artifacts, each could add different waveform signatures. After manual labeling, the wide array of possible artifacts are characterized into four types based on different signatures, i.e. sharpness of edge, duration and waveform shape. The four types of artifacts are shown in Fig. 3 and described as follows: Type-0 : It has dominant power spectrum in low frequency region. They may appear as a single waveform or in a periodic fash- ion in recordings and over different channels. For example, this artifact can be generated from the muscle activities of the subject during movement. Type-1 : It is similar to step response, where there is an instan- taneous large impulse whose effects are dampened over a period of time. Individual artifacts may have different widths and different decaying spectra with frequency. However, due to the sharp edge, there always exist localized high frequency features. The intercon- nect cable between the electrode and the pre-amplifier may work as antenna and picks-up some type-1 artifacts due to the subject’s motion. Type-2 : Type-2 artifacts have been observed more frequently compared with both Type-0 and Type 1 artifacts. They usually have two fast ramp edges at two ends and can be modeled as the deriva- tives of the type-1 artifacts, which suggest that type-2 artifacts are another form of type-1 artifacts possibly generated from the same sources. The settling of electrodes may also produce such type of artifacts. Type-3 : Type-3 artifacts are often large (sometimes cause recording saturation) and narrow in width (less than 200 s). They could appear both as individual waveforms and in a train, and tend to appear simultaneously over different channels. Different from neural spikes and local field potentials, they have a wide spectrum up to the low pass corner frequency set by recording electron- ics. This type of artifact may be generated from a sudden charge injection and discharging to the electrode/electronics. Apart from the mentioned four types, there could be other types of artifact present as well, but these four types are supposed to represent/cover most of the artifact types. Because the way they have been classified (i.e. based on waveform shapes, appearance and/or frequency characteristics) is quite universal and almost any appeared artifact we found can be fallen in one of the four types. This paper deals with the mentioned four types of artifacts for quantitative evaluation of the proposed algorithm. However, it is discussed later that the algorithm does not depend on the artifact types rather it depends on the spectrum characteristics of the signal of interest. Therefore it can be applied to other types of artifacts (beyond these four types) as well. Let’s assume r ( n ) as the recorded neural data at discrete-time instant n and it can be expressed ...