Application of machine learning algorithms in fMRI studies: a scientometric analysis

Background and aims: Machine learning is a subfield of artificial intelligence, a branch of datascience that allows computers to learn without explicit programming from existing training data.In recent years, there has been a significant increase in the application of various machine learningalgorithms in the context of functional magnetic resonance imaging (fMRI). It is critical todetermine which algorithms are applied in fMRI studies and how frequently they are employed.In order to provide a comprehensive roadmap for future research and strategic planning in thisfield, we performed a scientometric analysis of the scientific output of the application of machinelearning algorithms in fMRI.Method: In this scientometric study, a comprehensive search was conducted in Scopus using theterms “fMRI”, “functional magnetic resonance imaging”, “functional MRI”, “machine learning”,“deep learning”, “supervised learning”, “unsupervised learning”, and “reinforcement learning”up to August ۲۰۲۲. The publications that highlighted both fMRI and machine learning or indicatedthe application of a machine learning algorithm in fMRI data processing were included in thisstudy.Using the Bibliometrix package in the R ۴-۲-۱ programming language and the VOS viewer, theincluded documents were analyzed, and various scientometric parameters—such as the most-citedpublications or co-authorships—were reviewed.Results: A total of ۲۲۳۲ documents were obtained after the data were screened based on their titles,abstracts, and keywords. Since ۲۰۰۳, the studies have been published in ۵۳۰ distinct journals,with the most publications appearing in the journal “NEUROIMAGE.” From ۲۰۰۳ to ۲۰۲۲, thenumber of publications increased in this field of study. The top authors, as well as those with thehighest total link strength, were listed. The top author was cited ۱۰.۶۳ times per year out of the۷۷۳۵ authors that contributed to the publications. The United States was ranked among the topnations, having not only the most publications and citations but also the strongest cross-nationalinterconnection. The keywords in this context were also investigated, and “Machine learning”and “fMRI” were among the most popular themes in Density Visualizations. Furthermore, the topdocuments on basis of their citations were highlighted.Conclusion: The rising number of papers in recent years demonstrates a growing interest in theapplication of artificial intelligence in neuroimaging, necessitating additional research into themachine learning methods employed in fMRI studies. The results reveal that there is a need forgreater deep learning and classification model application among the various machine learning algorithms,and on the other hand, it seems that error prediction and reinforcement learning modelsin this field have been overused. The citation and connection between nations were allocated toadvanced countries, which might be attributed to the easier access to artificial intelligence technologiesas well as the increased emphasis on academic collaboration in these places. The resultsalso suggest that more machine-learning algorithms would be required in fMRI investigations ofschizophrenia patients and people suffering from depression.

کلیدواژه ها:

scientometrics ، functional magnetic resonance imaging (fMRI) ، Bibliometric package ، machine learning algorithms ، VOS viewer