The Bibliometric Analysis of EEGLAB Software in the Web of Science Indexed Articles

1. Introduction

The EEGLAB was presented as MATLAB (http://www.mathworks.com/) open-source toolbox in a research publication in 2004 by Arnaud Delorme and Scott Makeig from Swartz Center for Computational Neuroscience, Institute for Neural Computation, University of California San Diego, USA [1]. It has an interactive graphical User interface (GUI) with independent component analysis (ICA), Time/Frequency Analysis (TFA), and more than 150 plug-ins for example Fieldtrip-lite [2] , ERPLAB [3], ICLabel [4], SIFT [5] , AMICA [6], PACT [7] and LIMO [8] to analyze dynamic brain data. (https://sccn.ucsd.edu/eeglab/plugin_uploader/plugin_list_all.php)

Although it is not the only software in neuroscience research, it has a big community of users and developers from different countries and various expertise in neurosciences, biomechanics, Psychology, Bioengineering, Biosignal processing, Neuromechanics, Rehabilitation, Software engineering, Biostatistics, and data science. Also, it is used with other software for EEG- fMRI (Functional magnetic resonance imaging) datasets with SPM (Statistical Parametric Mapping) [9], EEG-NIRS (near-infrared spectroscopy) dataset with BBCI Toolbox [10], BCILAB in brain–computer interface (BCI) development [11], in R packages like neuroconductor [12] and medical researches [13] and the Virtual Brain (TVB) [14,15].

The systematic reviews and meta-analysis studies about EEG were highly cited and popular for example Default-mode brain dysfunction in mental disorders [16], deep learning [17], feature extraction [18] and meta-analysis for randomized controlled trials for Nonpharmacological interventions for ADHD [19]. But the bibliometric analysis is new and was limited to the application of EEG indices in human cognitive performance with 143 items [20], Mild Cognitive Impairment (MCI) research with 2310 items [21], mental fatigue on athletic performance with 658 items [22], Quantitative EEG in neuropsychiatric field with 1904 articles [23], neuromarketing with 30 items [24] and 24 items [25], Consumer Neurosciences with 364 items [26], consumer behavior and marketing with 497 items [27] , strategic management studies with 105 items [28] ,Neurorehabilitation with 874 items [29], Neuroarchitecture Assessment with 295 items [30] and Construction [31].

Sometimes bibliometric analysis is combined with different text mining methods such as topic modeling and word clouds. They show the most important words in a text by statistical and machine learning methods [32,33]. The recent study of AI-enhanced human EEG analysis with 2,053 research items presented world clouds [34]. In this study, we present the bibliometric analysis with some text mining methods for aggregated abstracts by using the bibliometrix R package for all available ISI research articles that have been cited the EEGLAB [35].

2. Materials and Methods

2.1. Data Gathering

The EEGLAB was introduced in the “Delorme A, Makeig S. EEGLAB: an open-source toolbox for analysis of single-trial EEG dynamics including independent component analysis. Journal of neuroscience methods. 2004 Mar 15;134(1):9-21.” [1]. Since 8/27/2023, there have been 20,464 citations in the Google Scholar. It consists of different types of articles, proceedings, poster presentations, etc. In this research, only available items in the Web of Science (WOS) Core Collection (2001-present) have been collected from webofknowledge.com. It consists of four databases: 1) Science Citation Index Expanded (SCI-EXPANDED)--2001-present, 2) Social Sciences Citation Index (SSCI)--2001-present, 3) Arts & Humanities Citation Index (AHCI)--2001-present and 4) Emerging Sources Citation Index (ESCI)--2018-present. The available items were 12,700 (~62.1% of all Google Scholar citations) and they were collected, integrated, and saved with bib format file. They were not all references such as only SCOPUS indexed journals, but only articles published by the Institute for Scientific Information (ISI) journals.

2.2. Data Analysis

The data and bibliography analysis were conducted with Bibliometrix [35] package in R studio 2023.06.1 and R Core Team (2022). [36]

3. Results

3.1. Descriptive Statistics

The timespan is from 2004 to 2023 with 12,700 documents published in 1,125 ISI-indexed sources (journals, books etc.), written by 29,125 authors, including 19,062 author’s keywords, 13,707 keywords PLUS and 279,617 references. The annual growth rate of publication is 28.12 %, international Co-authorship is 37.27 % and Co-authors per document is 4.89 and the average citations per document is 22.51. According to the Clarivate website, the keyword PLUS are words or phrases that frequently appear in the titles of an article's references, but do not appear in the title of the article itself.

3.2. Sources

The most relevant sources are Neuroimage, Frontiers in Human Neurosciences, Scientific Reports, Psychophysiology and PLOS One with 780, 526, 446,425 and 371 articles. According to the Bradford’s Law, these first five journals plus Journal of Neuroscience, Neuropsychologia, Clinical Neurophysiology, Frontiers in Neuroscience, Journal of Cognitive Neuroscience and International Journal of Psychophysiology have 4,318 (34.00%) articles and they are categorized as the core sources. These articles came from 11 out of 1,125 sources and they have 12,062 out of 29,125 authors. The local impact of the first five journals is presented at Table 1.

According to the definition of the Hirsch-index or H-index, “A scientist has index $h$ if $h$  of his or her $N_p$ papers have at least $h$  citations each and the other $(N_p-h)$  papers have fewer than $\le h$  citations each.” [37] The g-index is introduced as an improvement of the h-index to measure the global citation performance of a set of articles [38] It is the highest number $g$  of papers that together received $g^2$  or more citations . The M-Index definition is “ $\frac{h}{y}$  where $h=h-index$ , $y=number of years since publishing the first paper.$ ” [39]. According to Table 1, Neuroimage journal has the highest values of H-Index, G-Index ,and M-Index and total citations.

3.3. Authors

Some results of author analysis are not very reliable, because many author names have the same abbreviations especially in Chinese first and last names and their unique ORCID code is not available. Therefore, only related analysis was reported that the names are famous and related to the specific person.

The 19,416 (66.7%), 4,432 (15.2%), 1,879 (6.5%) and 292 (1.0 %) authors have only 1, 2, 3 and 7 articles, respectively. The most cited countries are the USA, Germany and the United Kingdom with 93,093, 32,621 and 20,748 total citations, respectively. The collaboration network between universities is estimated and clustered with Walktrap method into 6 clusters. [40] According to Figure 2, the biggest cluster is yellow with the University of California San Diego (UC) where the Swartz Center for Computational Neuroscience located, the hosting lab of EEGLAB. The red, green and brown clusters have only German, Chinese and European countries universities, respectively. The clusters also have relationships between each other. The university name and their countries are listed in Table 2.

We also estimate the collaborations between countries. In this regard, we only consider 50 first countries and put them into 3 clusters based on the Wlaktrap algorithm:

• Cluster 1: China, Japan, South Korea, Israel, India, Greece, Singapore, New Zealand, Malaysia, United Arab Emirates, Thailand, South Africa, Saudi Arabia, Pakistan, Bangladesh
• Cluster 2: USA, Germany, United Kingdom, Canada, Italy, France, Australia, Netherlands, Spain, Switzerland, Belgium, Finland, Denmark, Iran, Brazil, Norway, Hungary, Ireland, Poland, Austria, Portugal, Russia, Sweden, Turkey, Czech Republic, Lithuania, Mexico, Slovenia, Estonia, Serbia, Cuba, Luxembourg
• Cluster 3: Chile, Argentina, Colombia

3.4. Documents

According to Table 3, “Global Citations (TC) means the Total Citations that an article, included in your collection, has received from documents indexed on a bibliographic database (WoS, Scopus, etc.).”, the applications such as FieldTrip [2], Brainstrom [41], ERPLAB [3] and MNE-Python [42] have the highest total citations. [35]

According to Table 4, local citations are “the citations that a reference has received from documents included in your collection” [35], local to global ratio is above 50% for the ERPLAB app [3], ADJUST app [49], ICLabel app [4] and ICA and Blind Source Separation (BSS) [50].

According to Table 5, the most locally cited references are EEGLAB [1], FieldTrip[2] and Nonparametric statistical tests [54].

The Reference Publication Year Spectroscopy (RPYS) [61] is presented in the supplementary. The years before 1900 are omitted because the number of them is very neglect. The peak at 2004 is related to the [1] with about 32.4% of all 39,155 references in 2004. And the highest peaks is in 2012 and 2014 with 40,360 and 40,431 references, respectively. The decline in the graph shows after 2014.

The trend topic of keyword plus is plotted. (Figure 3) The dynamic, EEG and brain terms have the highest frequency in 2019, while safety, mini-mental state and attentional capture term have the highest frequency in 2022.

The thematic map [62,63] of trigram words in abstracts is presented in Figure 4. It has four parts: 1) Niche themes (low centrality and high density, limited importance) including Alternating Current Stimulation (tACS), Transcranial Current Stimulation (tDCS) and Rapid serial visual presentation (RSVP). 2) Emerging or declined themes (low centrality and low density, marginal) including Local Field Potential (LFP) and Deep brain stimulation. 3) Motor Themes (high centrality and high density, important for research) including Transcranial magnetic stimulation (TMS), Alzheimer diseases, mild cognitive impairment (MCI)and delta, theta, alpha. 4) Basic Themes (high centrality and low density, general topics) including Independent Component Analysis (ICA), Magnetic Resonance Imaging (MRI), Brain Computer Interface (BCI), Support Vector Machine (SVM), Event Related potential (ERP) and Mismatch negativity (MMN).

The co-citation network between authors shows the relationship between cited sources in the documents in two populations: 1) All sources in Figure 5 and Table 6 show three clusters. Dr. Arnaud Delorme is in the center of the authors. 2) Core sources based on the Bradford Law Zone (n = 4318) in Figure 6. It has 7 clusters with Dr. Scott Makeig and Dr. Arnaud Delorme in one cluster, Dr. Stefan Debener in other clusters and Dr. Mike X Cohen in another cluster. The other remaning clusters are shown in Figure 6.

4. Conclusions

Despite the emerging and the growth of open source Python and related MNE library [42] for computational neuroscience (with more than 2,000 Google citations ), EEGLAB has the highest number of google citations among similar software like SPM [64] (with more than 11,000 Google citations). Many global and famous universities and research institutes published research with EEGLAB in the USA, Europe, Canada, Japan, Australia and Russia. But it is not limited to developed countries, and many developing countries like China, India, Taiwan, Turkey, Iran, Saudi Arabia, Bangladesh, Brazil, Cuba, Argentina, Colombia, and many others use it in their scientific experiments and publications.

One of the main limitations of this research is that it only considers the ISI-indexed articles. Still, due to the large number of research articles, it covers many important aspects of literature. The second limitation is that it is not about all computational neuroscience papers, but it is only about the papers that cited the EEGLAB and with a high probability have EEG datasets. For example, tDSC and tACS have existed in the niche theme of Figure 4, but they are growing topics in the neuroscience literature. [65,66]. The one direction for future research is bibliographic analysis of special statistical methods with EEGLAB and EEG datasets for example, machine learning methods such as support vector machine [67], dimension reduction methods such as ICA [43] , functional data analysis methods [68,69], and deep learning methods [17,70].

Further analysis including world clouds, tree maps, bar charts of the most frequent words in keywords plus, keyword, title (unigram, bigram, trigram) and abstract (unigram, bigram, trigram) and many others are presented in the Supplementary Materials.