Epilepsy is a growing concern, especially in countries with aging populations. Epilepsy diagnosis remains a big challenge for medical practitioners. First, the diagnosis requires a multidisciplinary approach incorporating semiology and different neurophysiological and neuroimaging tests conducted using various methods like magnetoencephalography (MEG) and electroencephalography (EEG). The second aspect involves the visual interpretation of the MEG/EEG waveforms by specialists for accurate diagnosis of epilepsy. However, visual interpretation is bound to errors that might lead to inaccurate and unquantifiable diagnoses.
Automatic epilepsy diagnosis that does not depend on visual interpretation of MEG/EEG signals has recently attracted significant research attention. To this end, various features of the MEG/EEG signals contributing to the detection of seizures have been explored and applied to machine learning for possible automatic diagnosis. These features include functional connectivity, relative power, entropy and spike in the interictal state, which has all been proved to be beneficial in diagnosing epilepsy. It is hypothesized that epilepsy patients in the interictal state and healthy people may have different resting-state phase-amplitude coupling (PAC) capable of improving the discrimination between the groups. Nevertheless, the usefulness of interictal PAC in facilitating epilepsy diagnosis remains poorly understood.
Compared with machine learning methods, deep convolutional neural networks (DCNN) have been applied to many EEG/MEG signals to identify epileptic waveforms and discriminate patients from healthy participants with higher accuracy. This can be partly attributed to new electrophysiological features learned by DCNN though it is still difficult to understand what these new features are. Thus, improving automatic diagnosis requires applying machine learning to many patients with different forms of epilepsy. In addition, whether PAC in the interictal state is abnormal in epilepsy patients and its applicability in discriminating patients from healthy participants is also yet to be examined.
Herein to address these limitations, a team of researchers from Osaka University: Dr. Yuya Fujita, Professor Takufumi Yanagisawa, Specially Appointed Associate Professor Ryohei Fukuma, Natsuko Ura, Associate Professor Satoru Oshino and Professor Haruhiko Kishima characterized the abnormal PAC in the interictal state in epilepsy. They hypothesized that PAC in the resting state is abnormal in epilepsy patients in the interictal states and could improve the discrimination between two groups alongside other relevant features, including those extracted by deep learning. In their approach, cortical currents estimated from MRI and MEG were used to determine power in different bands. PAC was determined using synchronization index for eight frequent band pairs. Their work is currently published in the journal, Journal of Neural Engineering.
The research team showed that the mean synchronization index values were significantly different for healthy participants and patients with epilepsy. Specifically, the resting-state PAC and the synchronization index value difference were higher for θ/low γ in the temporal lobe. Consequently, the discrimination accuracy of the DCNN was significantly improved by PAC, allowing further high accuracy discrimination of the estimated cortical currents. For example, using a combination of DL and PAC, a maximum discrimination accuracy of 90% was attained. Though its underlying physiological mechanisms remain unknown, PAC was considered appropriate for this study owing to its contribution to different cortical functions and association with various neurological diseases like autism spectrum disorder and Alzheimer’s disease.
In summary, this is the first study to demonstrate abnormal PAC in patients in the interictal state. The combination of DCNN and physiologically characteristic features played an important role in improving the accuracy of automated epilepsy diagnosis. In a joint statement to Advances in Engineering, the authors explained their study will contribute to developing highly effective and accurate automated epilepsy diagnosis techniques.
Fujita, Y., Yanagisawa, T., Fukuma, R., Ura, N., Oshino, S., & Kishima, H. (2022). Abnormal phase–amplitude coupling characterizes the interictal state in epilepsy. Journal of Neural Engineering, 19(2), 026056.