A phonemic-rate rhythm, not speech
The study asked whether dyslexia can be detected from how the brain tracks a simple repeating rhythm. Forty-eight children around age 7 listened to white noise whose amplitude was modulated at three rates chosen to match Spanish speech units: 4.8 Hz for syllables, 16 Hz for intrasyllabic structure, and 40 Hz for phonemes. Each rate was played for 2.5 minutes in an ascending and descending sequence, for 15 minutes of recording in total.
The cohort came from the LEEDUCA project at the University of Málaga: 15 children with a clear reading deficit and 33 controls, matched on age and a school-level socioeconomic index. Because the stimulus contains no phonetic content, any group difference cannot be attributed to knowledge of letters or speech sounds. That design choice is what makes the result interesting: it points toward basic auditory timing rather than a speech-specific representation problem.
Coupling between stimulus phase and EEG amplitude
Entrainment here means that brain activity aligns with the rhythm of an external sound. The authors measured it with phase-amplitude coupling (PAC): the phase of the modulated stimulus was compared with the amplitude envelope of the EEG signal at each electrode, after filtering the EEG into narrow frequency bands. A strong PAC value means the EEG amplitude rises and falls in step with the stimulus rhythm.
Several PAC formulas were compared, including the Heights Ratio (HR), KL-MI, PLV, MVL and gcPAC. For each frequency band, the authors counted how many PAC values exceeded a permutation-derived Z-score threshold of 1.65, then normalized that count and averaged it across 5-second artifact-free segments. These band-level features were fed to a linear support vector classifier evaluated with repeated stratified 5-fold cross-validation across 31 EEG channels. The classifier's weights show which electrodes pushed a child toward the dyslexia or control label; they are markers of group separation, not direct measures of neural activity.
The 40 Hz Heights Ratio separates the groups best
The best combination was the 40 Hz stimulus with the gamma band, measured by the Heights Ratio. In the best case it reached a mean balanced accuracy of 0.929 (standard deviation 0.08 across cross-validation folds, p = 0.0069). Balanced accuracy averages sensitivity and specificity, so it is not inflated by the larger control group.
Performance depended strongly on both the stimulus rate and the PAC measure. The 16 Hz stimulus with HR reached about 80% (best balanced accuracy 0.830, p = 0.0225), and KL-MI at 16 Hz was slightly lower. For the 4.8 Hz stimulus, all measures performed worse, and the best result (0.738, p = 0.0945) was not statistically significant. The authors attribute the HR advantage to its sensitivity to large differences in the quantized amplitude distribution, whereas KL-MI measures divergence from uniformity and may respond less to the extreme modulation patterns that distinguish the groups.
Right-hemisphere coupling marks dyslexia, left marks control
The classifier weights for 40 Hz gamma entrainment were not symmetric across hemispheres. The largest positive weights, which pushed a child toward the dyslexia label, appeared at FC6 and P7, with a right-hemisphere line of FC6, C4 and CP6 contributing positively. Large negative weights, pushing toward the control label, appeared at left-hemisphere sites around language areas: P3, TP9, FC1 and F3, along with some right-hemisphere and midline electrodes.
In plain terms, stronger 40 Hz coupling over broad right-hemisphere regions went with a dyslexia classification, while stronger coupling at specific left-hemisphere language sites went with the control classification. The authors read this as a phonemic entrainment deficit in dyslexia, with the right-hemisphere pattern possibly reflecting compensation. They also note an alternative: atypical lateralization of auditory processing, in which the right-hemisphere pattern is a primary feature rather than a compensatory response. The classifier weights alone cannot decide between these explanations.
Right-hemisphere coupling tracks weaker reading scores
Among controls, the authors related aggregated HR-PAC features to seven behavioral indices using a Multi-task Lasso model with permutation testing and Benjamini–Hochberg correction across 31 channels and 7 indices. For 40 Hz gamma entrainment, higher coupling at C4 was associated with lower scores on pseudoword reading (coefficient −0.489, corrected p = 0.034), reading comprehension (−0.428, p = 0.040) and rapid automatized naming (−0.402, p = 0.016). FC6 showed negative coefficients of similar direction, though these did not survive correction.
Positive coefficients appeared at FC1 and Fz, near left premotor cortex, meaning stronger coupling there went with better scores. The authors interpret the right-hemisphere negative associations as consistent with a compensatory mechanism in dyslexia reported in earlier work, while cautioning that these are associations, not causal evidence. Note also that this regression was run in control participants, so it describes variation in reading performance within the control group rather than a dyslexia-specific mechanism.
A marker in one small cohort, not a diagnostic test
The headline accuracy is a best-case cross-validated estimate from 48 children, 15 of them with dyslexia, in a single Spanish cohort. The 0.929 figure carries a standard deviation of 0.08 across folds, and the authors describe it as the best case rather than a fixed operating point. A classifier trained on this sample would need independent validation before it could be treated as a screening tool.
The authors also note that the observed deficit appeared with non-linguistic stimuli, which suggests basic auditory temporal processing rather than speech-specific representations. That interpretation is theirs, and it is consistent with the temporal sampling framework they cite, but the study does not test whether auditory training would help. The practical claim supported here is narrower: 40 Hz amplitude-modulated noise, combined with Heights Ratio PAC over gamma-band EEG, carries group information that a linear classifier can exploit in this cohort.
The phonemic rate and the simpler measure carry the signal
The most decision-relevant comparison in this paper is not dyslexia versus control in general but which stimulus rate and which coupling measure produced the separation. The 40 Hz rate, chosen to match the average phonemic rate in Spanish, outperformed the syllable-rate 4.8 Hz stimulus, where no measure reached significance. Among measures, the Heights Ratio outperformed KL-MI, PLV, MVL and gcPAC, and the authors link this to how each formula responds to extreme amplitude modulation.
For readers weighing entrainment as a dyslexia marker, the priority is therefore specific: test fast, phoneme-rate rhythms and use a measure sensitive to large modulation differences, rather than assuming any rhythmic stimulus or any PAC formula will work. The right-hemisphere versus left-hemisphere split adds a second check, since the two patterns point in opposite directions and a single averaged coupling value would blur them.
Sources used in this answer
EEG Entrainment to Non-Speech Rhythmic Patterns Associated with Dimensions of Language Performance
Using 40 Hz amplitude-modulated white noise and Heights Ratio phase-amplitude coupling, a linear classifier separated dyslexic from control children with a best-case balanced accuracy of 0.929 (p = 0.0069) in 48 participants.