A super-resolution time–frequency representation method for differences between neural oscillations and its application on magnetoencephalography

Objective: It is paramount to detect neural oscillatory time–frequency (TF) differences under varied physiological conditions in the presence of noise and background activity in neuroscience. Therefore, TF analysis methods have been combined with statistical techniques for oscillatory difference TF representation (ODTFR). However, the detection ability of ODTFR is limited by the performance of TF analysis methods and statistical techniques. Thus, our objective was to propose a TF representation algorithm with high resolution and detectability. Methods: Superlet transform combined with cluster-depth-based permutation tests (SLT-CD) was proposed in this paper. SLT-CD was compared with commonly used algorithms. The detection score and TF bias were used to evaluate the algorithms using simulated oscillatory differences. Rényi entropy was used to evaluate the performance of the algorithms with Optically pumped magnetometer-magnetoencephalography (OPM-MEG) data. Results: Using the simulated data, SLT-CD showed better TF resolution and detection ability for oscillatory differences than those of other algorithms (p < 0.05) based on the detection score and TF bias. In analyzing data of two OPM-MEG experiments, SLT-CD exhibited the lowest Rényi entropy (on average, 30% and 60% lower than those of other algorithms), supporting the higher resolution and sensitivity. Conclusions: The proposed algorithm utilized the advantages of the SLT and cluster-depth-based permutation tests for the TF representation of oscillatory differences with high detectability and resolution. SLT-CD performed better than commonly used algorithms in both simulation and MEG experiments. Significance: Overall, the proposed method provides a new means for data-driven neural oscillation analyses under different stimuli or physiological states.