Fast, accurate, and interpretable decoding of electrocorticographic signals using dynamic mode decomposition.

Dynamic mode (DM) decomposition decomposes spatiotemporal signals into basic oscillatory components (DMs). DMs can improve the accuracy of neural decoding when used with the nonlinear Grassmann kernel, compared to conventional power features. However, such kernel-based machine learning algorithms have three limitations: large computational time preventing real-time application, incompatibility with non-kernel algorithms, and low interpretability. Here, we propose a mapping function corresponding to the Grassmann kernel that explicitly transforms DMs into spatial DM (sDM) features, which can be used in any machine learning algorithm. Using electrocorticographic signals recorded during various movement and visual perception tasks, the sDM features were shown to improve the decoding accuracy and computational time compared to conventional methods. Furthermore, the components of the sDM features informative for decoding showed similar characteristics to the high-γ power of the signals, but with higher trial-to-trial reproducibility. The proposed sDM features enable fast, accurate, and interpretable neural decoding.

Neural decoding of electrocorticographic signals using dynamic mode decomposition.

OBJECTIVE: Brain-computer interfaces (BCIs) using electrocorticographic (ECoG) signals have been developed to restore the communication function of severely paralyzed patients. However, the limited amount of information derived from ECoG signals hinders their clinical applications. We aimed to develop a method to decode ECoG signals using spatiotemporal patterns characterizing movement types to increase the amount of information gained from these signals. APPROACH: Previous studies have demonstrated that motor information could be decoded using powers of specific frequency bands of the ECoG signals estimated by fast Fourier transform (FFT) or wavelet analysis. However, because FFT is evaluated for each channel, the temporal and spatial patterns among channels are difficult to evaluate. Here, we used dynamic mode decomposition (DMD) to evaluate the spatiotemporal pattern of ECoG signals and evaluated the accuracy of motor decoding with the DMD modes. We used ECoG signals during three types of hand movements, which were recorded from 11 patients implanted with subdural electrodes. From the signals at the time of the movements, the modes and powers were evaluated by DMD and FFT and were decoded using support vector machine. We used the Grassmann kernel to evaluate the distance between modes estimated by DMD (DMD mode). In addition, we decoded the DMD modes, in which the phase components were shuffled, to compare the classification accuracy. MAIN RESULTS: The decoding accuracy using DMD modes was significantly better than that using FFT powers. The accuracy significantly decreased when the phases of the DMD mode were shuffled. Among the frequency bands, the DMD mode at approximately 100 Hz demonstrated the highest classification accuracy. SIGNIFICANCE: DMD successfully captured the spatiotemporal patterns characterizing the movement types and contributed to improving the decoding accuracy. This method can be applied to improve BCIs to help severely paralyzed patients communicate.