A Bayesian framework for phase-amplitude cross-frequency coupling inference: Application to reading disability detection

Abstract

Reading difficulties are often associated with altered brain connectivity, but detecting these differences reliably is challenging. We present a Bayesian phase-amplitude coupling (PAC) framework to measure cross-frequency brain interactions, addressing the limitations of traditional PAC methods in EEG. Unlike standard PAC approaches that may miss complex directional interactions between brain rhythms, our Bayesian model incorporates prior knowledge of significant coupling at each electrode to guide its estimations, yielding a robust measure of neural synchronization both within and across brain regions. We applied this model to EEG recordings from 48 children (15 with reading difficulties, 33 controls) during auditory steady-state stimulation at 4.8, 16, and 40 Hz. The Bayesian approach revealed clear cross-frequency coupling patterns: significant theta–gamma coupling was found in both groups, especially in occipital–parietal regions involved in phonological processing and attention. Importantly, the reading difficulties group showed stronger and more widespread frontoparietal coupling at 16 Hz than the controls, including a prominent connection from electrode CP6 to FC6-suggesting a possible compensatory mechanism or disrupted pathway. No significant coupling was detected at 40 Hz, though near-significant trends hint at a subtle role for gamma oscillations. Finally, using PAC features from our model, a simple classifier distinguished children with and without reading difficulties with balanced accuracies around 75–80 % (significantly above chance), demonstrating the method’s practical efficacy. These results highlight that the Bayesian PAC framework not only uncovers meaningful brain connectivity patterns in noisy EEG data but also serves as a promising tool for identifying biomarkers of reading disabilities and potentially other cognitive conditions.