ML and DL Models for Stroke Prediction from Bio-Signals: A Systematic Review and Bibliometric Analysis

Abstract

Strokes continue to be a primary reason for disability and death around the globe. Annually, over 12.2 million new strokes occur, which necessitates the development of early detection and intervention tools to reduce the potential harm. This systematic review and bibliometric analysis aim to review and visualize recent advances in predicting stroke or post-stroke effects using bio-signals, either with machine learning (ML) or deep learning (DL). The included studies were published between 2016 and 2024. A comprehensive search of IEEE, PubMed, MDPI, and ScienceDirect databases was performed using keywords related to stroke prediction, machine learning, deep learning, and bio-signals. From an initial pool of 152 studies, 15 studies met the inclusion criteria through the screening process. South Korea contributed the most to publishing studies on stroke prediction using bio-signals. The results show that Electroencephalography (EEG) is the most used bio-signal in the reviewed studies. The sample size ranged from 3 to 4068. The top ten cited journals in the selected literature are high-ranked journals, which indicates the scientific validity of the concept and its potential for dissemination.