Xinshan Zhu

Bio: Xinshan Zhu is an academic researcher from Tianjin University. The author has contributed to research in topics: Compressed sensing & Weighting. The author has an hindex of 1, co-authored 1 publications receiving 12 citations.

Training-free compressed sensing for wireless neural recording using analysis model and group weighted [Formula: see text]-minimization.

Abstract: OBJECTIVE Data compression is crucial for resource-constrained wireless neural recording applications with limited data bandwidth, and compressed sensing (CS) theory has successfully demonstrated its potential in neural recording applications. In this paper, an analytical, training-free CS recovery method, termed group weighted analysis [Formula: see text]-minimization (GWALM), is proposed for wireless neural recording. APPROACH The GWALM method consists of three parts: (1) the analysis model is adopted to enforce sparsity of the neural signals, therefore overcoming the drawbacks of conventional synthesis models and enhancing the recovery performance. (2) A multi-fractional-order difference matrix is constructed as the analysis operator, thus avoiding the dictionary learning procedure and reducing the need for previously acquired data and computational complexities. (3) By exploiting the statistical properties of the analysis coefficients, a group weighting approach is developed to enhance the performance of analysis [Formula: see text]-minimization. MAIN RESULTS Experimental results on synthetic and real datasets reveal that the proposed approach outperforms state-of-the-art CS-based methods in terms of both spike recovery quality and classification accuracy. SIGNIFICANCE Energy and area efficiency of the GWALM make it an ideal candidate for resource-constrained, large scale wireless neural recording applications. The training-free feature of the GWALM further improves its robustness to spike shape variation, thus making it more practical for long term wireless neural recording.

Deep compressive autoencoder for action potential compression in large-scale neural recording.

Abstract: Objective Understanding the coordinated activity underlying brain computations requires large-scale, simultaneous recordings from distributed neuronal structures at a cellular-level resolution. One major hurdle to design high-bandwidth, high-precision, large-scale neural interfaces lies in the formidable data streams (tens to hundreds of Gbps) that are generated by the recorder chip and need to be online transferred to a remote computer. The data rates can require hundreds to thousands of I/O pads on the recorder chip and power consumption on the order of Watts for data streaming alone. One of the solutions is to reduce the bandwidth of neural signals before transmission. Approach We developed a deep learning-based compression model to reduce the data rate of multichannel action potentials. The proposed compression model is built upon a deep compressive autoencoder (CAE) with discrete latent embeddings. The encoder network of CAE is equipped with residual transformations to extract representative features from spikes, which are mapped into the latent embedding space and updated via vector quantization (VQ). The indexes of VQ codebook are further entropy coded as the compressed signals. The decoder network reconstructs spike waveforms with high quality from the quantized latent embeddings through stacked deconvolution. Main results Extensive experimental results on both synthetic and in vivo datasets show that the proposed model consistently outperforms conventional methods that utilize hand-crafted features and/or signal-agnostic transformations and compressive sensing by achieving much higher compression ratios (20-500×) and better or comparable reconstruction accuracies. Testing results also indicate that CAE is robust against a diverse range of imperfections, such as waveform variation and spike misalignment, and has minor influence on spike sorting accuracy. Furthermore, we have estimated the hardware cost and real-time performance of CAE and shown that it could support thousands of recording channels simultaneously without excessive power/heat dissipation. Significance The proposed model can reduce the required data transmission bandwidth in large-scale recording experiments and maintain good signal qualities, which will be helpful to design power-efficient and lightweight wireless neural interfaces. We have open sourced the code implementation of the work at https://github.com/tong-wu-umn/spike-compression-autoencoder.

EEG Motor Imagery Classification With Sparse Spectrotemporal Decomposition and Deep Learning

Abstract: Classification of electroencephalogram-based motor imagery (MI-EEG) tasks raises a big challenge in the design and development of brain–computer interfaces (BCIs). In view of the characteristics of nonstationarity, time-variability, and individual diversity of EEG signals, a deep learning framework termed SSD-SE-convolutional neural network (CNN) is proposed for MI-EEG classification. The framework consists of three parts: 1) the sparse spectrotemporal decomposition (SSD) algorithm is proposed for feature extraction, overcoming the drawbacks of conventional time–frequency analysis methods and enhancing the robustness to noise; 2) a CNN is constructed to fully exploit the time–frequency features, thus outperforming traditional classification methods both in terms of accuracy and kappa value; and 3) the squeeze-and-excitation (SE) blocks are adopted to adaptively recalibrate channelwise feature responses, which further improves the overall performance and offers a compelling classification solution for MI-EEG applications. Experimental results on two datasets reveal that the proposed framework outperforms state-of-the-art methods in terms of both classification quality and robustness. The advantages of SSD-SE-CNN include high accuracy, high efficiency, and robustness to cross-trial and cross-session variations, making it an ideal candidate for long-term MI-EEG applications. Note to Practitioners —Motor imagery-based brain–computer interfaces (MI-BCIs) are widely used to allow a user to control a device using only his or her neural activity. This article proposed a new framework to classify two-class MI tasks based on electroencephalography (EEG) signals. In this framework, a new sparse spectrotemporal decomposition method is used to extract time–frequency features from EEG signals. A convolutional neural network with squeeze-and-excitation blocks is then constructed to classify the MI tasks. We show the superiority of our method on two datasets and prove its feasibility for long-term MI-BCI applications.

Efficient Compressed Sensing for Wireless Neural Recording: A Deep Learning Approach

Abstract: Data compression is crucial for resource-constrained wireless neural recording applications with limited data bandwidth, and compressed sensing (CS) theory has successfully demonstrated its potential in neural recording applications. Based on deep learning theory, this paper presents a binarized autoencoder scheme for CS, in which a binary sensing matrix and a noniterative recovery solver are jointly optimized. Experimental results on synthetic dataset reveal that the proposed approach outperforms the state-of-the-art CS-based methods both in terms of recovery quality and computation time.