BCIwiki (bciwiki.com) — A study based on brain-computer interface (BCI) technology has proposed a personalized paradigm and decoding method for objective auditory frequency difference limen (FDL) evaluation, validated in 11 healthy participants. The research, conducted by scientists including those from Tianjin University, was published in Progress in Biochemistry and Biophysics on July 24, 2026. Traditional FDL measurement relies on active behavioral responses and is susceptible to subjective factors; existing paradigms often use uniform stimulus configurations that overlook individual perceptual differences. The study designed a personalized rapid serial auditory presentation (RSAP) paradigm based on each individual's precise FDL, using pure-tone sequences at 4,000 Hz to evaluate participants, simulating a realistic auditory environment to investigate neural representations of weak frequency deviations at the threshold state.
Given the complex spatiotemporal distribution of auditory evoked response features across multiple frequency domains, the researchers proposed the Multi-Scale Spatial-Temporal Dual Attention Network (MS-STAMNet). The network constructs parallel processing branches with multiple receptive fields, introduces a dynamic adaptive weighting strategy to localize core neural activity, and integrates multi-scale information through cross-branch feature interaction, achieving robust single-trial decoding of weak auditory evoked responses. Electrophysiological data analysis showed that subtle frequency deviation stimuli at threshold level successfully elicited pronounced N2 and P3 event-related potentials distributed over frontal, central, and temporal regions; in the time-frequency domain, event-related synchronization appeared in low-frequency delta and theta bands, accompanied by widespread event-related desynchronization in the higher alpha band. Model performance comparison showed MS-STAMNet achieved an average unweighted average recall (UAR) of (69.67±6.12)% and area under the curve (AUC) of 0.7618±0.07, significantly outperforming baseline models such as EEGNet and PLNet. Regression analysis (R²=0.016, P=0.709) verified a dissociation between neural decoding and behavioral performance, indicating the model captures implicit features of subtle frequency deviations even when they fail to trigger explicit conscious responses. Attention weight visualization further revealed the network's precise focus on key features over bilateral temporal and fronto-parietal regions.
The study concludes that it systematically reveals the multi-dimensional spatiotemporal evolution of neural responses to subtle acoustic variations under long-sequence threshold auditory stimulation, validates the efficacy and robustness of MS-STAMNet in decoding weak single-trial EEG signals amidst complex background noise, and lays a theoretical and methodological foundation for objective and quantitative evaluation of individual auditory cognitive abilities in clinical applications.