Bimodal EEG-fNIRS Models Outperform Unimodal Methods in Auditory Analysis

The Bottom Line

Precision requires complete data. This study establishes that bimodal EEG-fNIRS integration drastically improves the detection of cortical auditory evoked responses. By fusing electrical and hemodynamic signals, researchers achieved classification accuracy unattainable by EEG alone.

The Intelligence Gap and Bimodal EEG-fNIRS

Clinical audiology typically relies on electroencephalography (EEG). It is the standard. It captures the rapid electrical firing of neurons. However, EEG lacks spatial precision and provides no data on metabolic consumption. It sees the spark but misses the fuel. Functional near-infrared spectroscopy (fNIRS) fills this void by measuring cerebral haemodynamics. The objective of this research was to determine if fusing these distinct data streams could better classify auditory responses across five intensity levels.

Methodological Architecture

The investigators developed two primary models to process the data. The TS-model employed a convolutional neural network (CNN) fed by raw time-series data. In contrast, the F-model used a multi-layer perceptron (MLP) dependent on extracted features. These deep learning architectures were tested against three conventional machine learning classifiers. The critical comparison lay between unimodal EEG inputs and the fused bimodal dataset.

Mechanism: The Superiority of Fusion

The results provide a definitive tactical advantage for the combined approach. The TS-model, leveraging raw data, proved most effective, suggesting intrinsic patterns in the time-series are more predictive than manual features. The performance metrics were stark:

• Accuracy: Bimodal input hit 92.2%, compared to 79.3% for EEG alone.
• Robustness: The AUC metric improved to 94.4% from 80.5%.
• Consistency: F1-scores rose from 77.5% to 89.6%.

The addition of fNIRS data creates a composite signal. The algorithm utilises the slow haemodynamic response to validate and contextualise the rapid EEG spikes. This cross-verification eliminates ambiguity in detecting intensity-dependent responses.

Strategic Implications

The study measured algorithmic performance, but the implications extend to clinical practice. The substantial accuracy gap suggests that current unimodal methods may fail to capture the full extent of cortical processing. For patients unable to provide behavioural feedback, such as infants or those with severe neurological deficits, this is vital. A bimodal system offers a higher probability of correct diagnosis. It moves the field towards objective, automated hearing assessments that track not just the reception of sound, but the brain's metabolic engagement with it. This is a step towards high-fidelity neuro-monitoring.