Controlled Evaluation of EEG Foundation Models: Random Initialisation, Band-Power and Validation-Selected Probes

Pretrained EEG foundation models are proposed for clinical decoding, but whether reported gains transfer across populations or survive negative controls is unclear. We benchmark LaBraM, EEGMamba, CBraMod, REVE, LEAD, BENDR, and BIOT on five clinical tasks across four datasets. Primary analyses use frozen linear probes with subject-disjoint LOSO or grouped five-fold validation. Because CAUEEG releases no patient identifiers, it is evaluated at recording level with a patient-disjoint sensitivity. We challenge apparent gains using stronger classical comparators, label permutation, scrambled-label fine-tuning, and random-initialisation controls. In a matched 19-channel CAUEEG evaluation (Normal/MCI/Dementia; N = 1,187 recordings), classical features achieve 0.734 macro-AUROC versus 0.699 for BIOT, 0.669 for CBraMod, and 0.568 for REVE. A patient-disjoint sensitivity retains the classical-over-REVE ordering (0.717 versus 0.565). Dataset identity is decoded from frozen REVE embeddings at or near ceiling across Western-Korean and Western-Western pairs, including after PCA-50 and removal of line-frequency and amplitude-scale information. This establishes dataset membership, not a causal site or population effect. A matched random-initialised encoder exceeds pretrained REVE on CAUEEG (0.659 versus 0.570). On CHB-MIT cross-subject ictal detection (n = 23), REVE reaches 0.793, versus 0.739 for the best enhanced nonlinear comparator, 0.701 for random initialisation, and 0.505 for raw-signal random features. Because preprocessing removes absolute amplitude, this does not establish superiority over every plausible handcrafted baseline. Conclusions change materially after montage matching, patient-overlap checks, stronger comparators, and representation controls. We distill these checks into a reporting protocol for clinical EEG foundation-model studies.

EEG foundation models are advancing at an unprecedented pace, yet their evaluation is driven by downstream accuracy, which reveals little about whether their predictions rely on physiologically interpretable signal structure. We introduce a set of perturbation probes in two grounding axes: spatial grounding in channel identity and topographic locality, and spectral grounding in oscillatory content and waveform morphology. We apply these probes to five published EEG foundation models across motor-imagery and seizure recordings. Our experiments reveal differences in both representational sensitivity and downstream reliance. Electrode identity may be encoded without influencing downstream classification: permuting LaBraM's positional codes heavily displaced its embeddings yet reduced motor-imagery accuracy by only 1.2 points, while spectral details may be discriminable without producing specific responses to simulated clinical markers, as only REVE exceeded a spectral baseline on this task. These dissociations are hidden in common benchmarks, demonstrating how grounding probes can complement performance-based evaluation by revealing what EEG foundation models encode and what their predictions depend on.

Evaluating foundation models under appropriate adaptation settings is essential for understanding the quality and transferability of the learned representations. Recent EEG foundation models have demonstrated promising transfer capabilities across tasks and datasets, motivating their growing use in neurotechnology and clinical applications. However, these models are typically evaluated under full fine-tuning on well-curated downstream datasets, a setting that does not reflect biomedical domain constraints such as limited labeled data, reduced sensor coverage, or parameter-efficient adaptation. In this work, we propose a multi-dimensional evaluation framework for assessing EEG models under realistic low-resource conditions. Empirical analysis of both supervised EEG models and recent EEG foundation models, including LaBraM, CSBrain, and CBraMod, across 6 different datasets is performed under the proposed multi-dimensional evaluation framework. We find that EEG foundation models consistently provide performance gains on long-context tasks such as sleep stage prediction and mental health state classification. In contrast, for short-window Brain Computer Interface style tasks, supervised models achieve comparable despite having substantially fewer parameters. Additional analyses demonstrate that current foundation models provide limited robustness to short-window tasks and channel constrained settings. Together, these findings motivate the use of multi-dimensional evaluation protocols that characterize model behavior under realistic use constraints.