Investigating the Single Trial Detectability of Cognitive Face Processing by a Passive Brain-Computer Interface

Pham Xuan, Rebecca; Andreessen, Lena M.; Zander, Thorsten O. · 2022 · Crossref

DOI: 10.3389/fnrgo.2021.754472

archive: archived pipeline: cataloged

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study investigates whether a passive Brain-Computer Interface (pBCI) can detect the cognitive processing of faces in a single trial, aiming to support autonomous driving systems by interpreting non-verbal communication from pedestrians or drivers. The motivation stems from the limitation of current automated face recognition, which struggles with complex real-world scenarios and cannot infer intent or cognitive states. By utilizing the driver’s brain as a sensor, a pBCI could implicitly identify when a driver is processing facial mimicry, thereby enhancing the vehicle’s context awareness. The core research question is whether EEG responses specific to face recognition, particularly in the fusiform gyrus, can be reliably distinguished from responses to other visual stimuli in a single-trial classification. The experimental design involved a laboratory study with 11 participants (aged 24–34) who viewed three categories of stimuli: faces, houses (concrete objects), and abstract images (evoking "maybe face" responses). EEG data were recorded using 64 channels, with preprocessing including Independent Component Analysis (ICA) to isolate cortical components and remove artifacts such as eye blinks. The pBCI was calibrated using regularized discriminant analysis (LDA) to classify single trials based on features extracted from 50 ms windows starting 200 ms after stimulus onset. The classification performance was evaluated using 10-fold cross-validation, and source localization was performed to identify the neural origins of the discriminative signals. The results demonstrated that the pBCI could distinguish brain responses to faces from those evoked by houses or abstract stimuli with an accuracy exceeding 70% in a single trial. Specifically, the misclassification rate for distinguishing faces from non-faces (houses) was 27.75% when eye components were removed, indicating better-than-random performance. Source analysis identified activation patterns in the fusiform gyrus (BA 18) and the cingulate cortex that corresponded to face recognition. The N170 component, a known marker of face processing, showed stronger amplitudes for face stimuli compared to non-face stimuli in specific cortical clusters. However, post-hoc statistical tests on the N170 and P1 components in the fusiform gyrus did not reach significance, suggesting that while the classifier is effective, the underlying neural mechanisms may involve distributed networks rather than isolated peaks. The significance of this work lies in demonstrating the feasibility of using passive BCIs to decode high-level cognitive processes like face recognition in real-time. The findings suggest that the driver’s implicit brain activity can serve as a sensor for autonomous vehicles, potentially allowing cars to interpret non-verbal cues that are otherwise inaccessible to computer vision systems. While the study establishes a proof-of-concept in a controlled laboratory setting, the authors note that future research must address the transfer of this technology to real-world driving environments and its integration into artificial intelligence systems for autonomous driving. The study highlights the potential of neuroadaptive human-computer interaction to bridge the gap between machine perception and human cognitive understanding.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.8-27b-gittensor on 2026-08-23; verification: pending re-verification.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• neuro workload indices