The application of traditional machine learning and deep learning with EEG for mild cognitive impairment: a bibliometric analysis.

Abstract

This study uses bibliometric methods to analyze the status of electroencephalography (EEG) and machine learning applications in mild cognitive impairment (MCI) research. Relevant literature was searched in the four major English databases (Web of Science, PubMed, IEEE Xplore, and Scopus) and the three major Chinese databases (CNKI, Wanfang, and VIP) from their inception to June 2026. After removing duplicates using NoteExpress and screening the literature according to the inclusion and exclusion criteria, the final data were imported into CiteSpace 6.4. R2 and VOSviewer 1.6.20 for analysis of publication trends, geographic distribution, highly co-cited references, keyword clustering, and burst detection. A total of 547 valid studies were included in the analysis, comprising 523 English-language studies and 24 Chinese core journal articles. The number of publications in this field has shown phased growth. It entered a period of rapid development after 2018. China ranks first globally in terms of English-language publications; however, there is a significant gap between the number of core publications in Chinese and English. <i>Clinical Neurophysiology</i> is the journal with the highest total local citation score (TLS). Four of the top 10 highly co-cited journals are published in the United States. Highly cited literature has jointly contributed to establishing a core body of knowledge in this field. This knowledge encompasses clinical diagnostic guidelines, electrophysiological mechanisms, signal processing methods, and intelligent algorithm models. Traditional machine learning keywords focus on methodological exploration, including feature extraction and support vector machine classification. Deep learning keywords focus on model applications, including convolutional and artificial neural networks. Keyword burst analysis identifies feature selection, decision trees, and neuropsychological assessment as the current frontier topics in this field. Electroencephalography combined with machine learning has become an important research direction for the early identification and assessment of MCI. The research focus has shifted from traditional EEG signal analysis and manual feature engineering to optimization of deep learning models and exploration of multimodal data fusion. Improved algorithm interpretability, refined EEG rhythm and brain region localization, and the combination of neuropsychological assessment and EEG indicators have become a new research trend.