Human-Centric Cognitive State Recognition Using Physiological Signals: A Systematic Review of Machine Learning...

Area of Science:

• Neuroscience
• Computer Science
• Biomedical Engineering

Background:

• Cognitive state recognition is crucial for enhancing performance and safety in high-stakes environments.
• Advancements in analyzing physiological signals have been driven by machine learning (ML) and deep learning (DL).

Purpose of the Study:

• To systematically review advancements in cognitive state recognition from 2010 to early 2024.
• To analyze the evolution of methodologies, data modalities, and applications in the field.

Main Methods:

• Systematic review of 405 articles from five major databases (Scopus, Engineering Village, Web of Science, IEEE Xplore, PubMed).
• Inclusion criteria focused on studies using physiological signals and ML/DL in practical settings.
• Analysis of trends in methodologies (ML vs. DL), data modalities (neuroimaging vs. multimodal), and application domains.

Main Results:

• A significant shift from traditional ML to DL for analyzing physiological signals, with DL dominating by 2023.
• Increased use of multimodal physiological signals over neuroimaging for more comprehensive insights.
• Electrocardiogram (ECG) emerged as the most utilized modality, often analyzed with convolutional neural networks (CNNs).

Conclusions:

• Deep learning and multimodal physiological signal integration, particularly ECG with CNNs, represent the current state-of-the-art in cognitive state recognition.
• Future research trends include privacy-preserving methods like differential privacy and federated learning.
• Cognitive state recognition holds significant potential for enhancing performance, safety, and innovation across automotive, aviation, maritime, and healthcare sectors.