A large-scale neural network training framework for generalized estimation of single-trial population dynamics.

Achieving state-of-the-art performance with deep neural population dynamics models requires extensive hyperparameter tuning for each dataset. AutoLFADS is a model-tuning framework that automatically produces high-performing autoencoding models on data from a variety of brain areas and tasks, without behavioral or task information. We demonstrate its broad applicability on several rhesus macaque datasets: from motor cortex during free-paced reaching, somatosensory cortex during reaching with perturbations, and dorsomedial frontal cortex during a cognitive timing task.

Estimating muscle activation from EMG using deep learning-based dynamical systems models.

Objective. To study the neural control of movement, it is often necessary to estimate how muscles are activated across a variety of behavioral conditions. One approach is to try extracting the underlying neural command signal to muscles by applying latent variable modeling methods to electromyographic (EMG) recordings. However, estimating the latent command signal that underlies muscle activation is challenging due to its complex relation with recorded EMG signals. Common approaches estimate each muscle's activation independently or require manual tuning of model hyperparameters to preserve behaviorally-relevant features.Approach. Here, we adapted AutoLFADS, a large-scale, unsupervised deep learning approach originally designed to de-noise cortical spiking data, to estimate muscle activation from multi-muscle EMG signals. AutoLFADS uses recurrent neural networks to model the spatial and temporal regularities that underlie multi-muscle activation.Main results. We first tested AutoLFADS on muscle activity from the rat hindlimb during locomotion and found that it dynamically adjusts its frequency response characteristics across different phases of behavior. The model produced single-trial estimates of muscle activation that improved prediction of joint kinematics as compared to low-pass or Bayesian filtering. We also applied AutoLFADS to monkey forearm muscle activity recorded during an isometric wrist force task. AutoLFADS uncovered previously uncharacterized high-frequency oscillations in the EMG that enhanced the correlation with measured force. The AutoLFADS-inferred estimates of muscle activation were also more closely correlated with simultaneously-recorded motor cortical activity than were other tested approaches.Significance.This method leverages dynamical systems modeling and artificial neural networks to provide estimates of muscle activation for multiple muscles. Ultimately, the approach can be used for further studies of multi-muscle coordination and its control by upstream brain areas, and for improving brain-machine interfaces that rely on myoelectric control signals.

Predicting the outcome of laser peripheral iridotomy for primary angle closure suspect eyes using anterior segment optical coherence tomography.

PURPOSE: Develop an algorithm to predict the success of laser peripheral iridotomy (LPI) in primary angle closure suspect (PACS), using pretreatment anterior segment optical coherence tomography (ASOCT) scans. METHODS: A total of 69 eyes with PACS underwent LPI and time-domain ASOCT scans (temporal and nasal cuts) were performed before and after LPI. After LPI, success is defined as one or more angles changed from closed to open. All the pretreatment ASOCT scans were analysed using the Anterior Segment Analysis Program to derive anterior chamber angle (ACA) measurements. The measurements for each angle were ordered along with angle-independent measurements totalling to 42 measurements which serve as features for the prediction algorithm. Two masked glaucoma fellowship-trained ophthalmologists graded the pre-LPI ASOCT scans to determine whether LPI was likely to successful. RESULTS: There were 42 (60.9%) eyes that fulfilled the criteria for success after LPI. Iris concavity, angle recess area (750 µm) and iris concavity ratio showed the highest predictive score and were selected using correlation-based subset selection method. These features were classified into two ('successful' and 'unsuccessful') categories using a Bayes classifier. The algorithm predicted the success of LPI with 79.28% cross validation accuracy, which was superior to the predictive accuracy of the ophthalmologists (kappa 0.497 and 0.636 respectively). CONCLUSION: Using pretreatment ASOCT scans, our algorithm was superior to ophthalmologists in predicting the success of LPI for PACS eyes. This novel algorithm could aid decision making in offering LPI as a prophylaxis for PACS.

Noise-robust unsupervised spike sorting based on discriminative subspace learning with outlier handling.

OBJECTIVE: Spike sorting is a fundamental preprocessing step for many neuroscience studies which rely on the analysis of spike trains. Most of the feature extraction and dimensionality reduction techniques that have been used for spike sorting give a projection subspace which is not necessarily the most discriminative one. Therefore, the clusters which appear inherently separable in some discriminative subspace may overlap if projected using conventional feature extraction approaches leading to a poor sorting accuracy especially when the noise level is high. In this paper, we propose a noise-robust and unsupervised spike sorting algorithm based on learning discriminative spike features for clustering. APPROACH: The proposed algorithm uses discriminative subspace learning to extract low dimensional and most discriminative features from the spike waveforms and perform clustering with automatic detection of the number of the clusters. The core part of the algorithm involves iterative subspace selection using linear discriminant analysis and clustering using Gaussian mixture model with outlier detection. A statistical test in the discriminative subspace is proposed to automatically detect the number of the clusters. MAIN RESULTS: Comparative results on publicly available simulated and real in vivo datasets demonstrate that our algorithm achieves substantially improved cluster distinction leading to higher sorting accuracy and more reliable detection of clusters which are highly overlapping and not detectable using conventional feature extraction techniques such as principal component analysis or wavelets. SIGNIFICANCE: By providing more accurate information about the activity of more number of individual neurons with high robustness to neural noise and outliers, the proposed unsupervised spike sorting algorithm facilitates more detailed and accurate analysis of single- and multi-unit activities in neuroscience and brain machine interface studies.

A fast, robust algorithm for power line interference cancellation in neural recording.

OBJECTIVE: Power line interference may severely corrupt neural recordings at 50/60 Hz and harmonic frequencies. The interference is usually non-stationary and can vary in frequency, amplitude and phase. To retrieve the gamma-band oscillations at the contaminated frequencies, it is desired to remove the interference without compromising the actual neural signals at the interference frequency bands. In this paper, we present a robust and computationally efficient algorithm for removing power line interference from neural recordings. APPROACH: The algorithm includes four steps. First, an adaptive notch filter is used to estimate the fundamental frequency of the interference. Subsequently, based on the estimated frequency, harmonics are generated by using discrete-time oscillators, and then the amplitude and phase of each harmonic are estimated by using a modified recursive least squares algorithm. Finally, the estimated interference is subtracted from the recorded data. MAIN RESULTS: The algorithm does not require any reference signal, and can track the frequency, phase and amplitude of each harmonic. When benchmarked with other popular approaches, our algorithm performs better in terms of noise immunity, convergence speed and output signal-to-noise ratio (SNR). While minimally affecting the signal bands of interest, the algorithm consistently yields fast convergence (<100 ms) and substantial interference rejection (output SNR >30 dB) in different conditions of interference strengths (input SNR from -30 to 30 dB), power line frequencies (45-65 Hz) and phase and amplitude drifts. In addition, the algorithm features a straightforward parameter adjustment since the parameters are independent of the input SNR, input signal power and the sampling rate. A hardware prototype was fabricated in a 65 nm CMOS process and tested. Software implementation of the algorithm has been made available for open access at https://github.com/mrezak/removePLI. SIGNIFICANCE: The proposed algorithm features a highly robust operation, fast adaptation to interference variations, significant SNR improvement, low computational complexity and memory requirement and straightforward parameter adjustment. These features render the algorithm suitable for wearable and implantable sensor applications, where reliable and real-time cancellation of the interference is desired.

Unsupervised spike sorting based on discriminative subspace learning.

Spike sorting is a fundamental preprocessing step for many neuroscience studies which rely on the analysis of spike trains. In this paper, we present two unsupervised spike sorting algorithms based on discriminative subspace learning. The first algorithm simultaneously learns the discriminative feature subspace and performs clustering. It uses histogram of features in the most discriminative projection to detect the number of neurons. The second algorithm performs hierarchical divisive clustering that learns a discriminative 1-dimensional subspace for clustering in each level of the hierarchy until achieving almost unimodal distribution in the subspace. The algorithms are tested on synthetic and in-vivo data, and are compared against two widely used spike sorting methods. The comparative results demonstrate that our spike sorting methods can achieve substantially higher accuracy in lower dimensional feature space, and they are highly robust to noise. Moreover, they provide significantly better cluster separability in the learned subspace than in the subspace obtained by principal component analysis or wavelet transform.

A new EC-PC threshold estimation method for in vivo neural spike detection.

This paper models in vivo neural signals and noise for extracellular spike detection. Although the recorded data approximately follow Gaussian distribution, they clearly deviate from white Gaussian noise due to neuronal synchronization and sparse distribution of spike energy. Our study predicts the coexistence of two components embedded in neural data dynamics, one in the exponential form (noise) and the other in the power form (neural spikes). The prediction of the two components has been confirmed in experiments of in vivo sequences recorded from the hippocampus, cortex surface, and spinal cord; both acute and long-term recordings; and sleep and awake states. These two components are further used as references for threshold estimation. Different from the conventional wisdom of setting a threshold at 3×RMS, the estimated threshold exhibits a significant variation. When our algorithm was tested on synthesized sequences with a different signal to noise ratio and on/off firing dynamics, inferred threshold statistics track the benchmarks well. We envision that this work may be applied to a wide range of experiments as a front-end data analysis tool.

Power line interference cancellation in in-vivo neural recording.

This paper presents an algorithm for removing power line interference in neural recording experiments. It does not require any interference reference signal and can reliably track interference changes in frequency, phase, and amplitude. The method includes three major steps. First, it employs a robust frequency estimator to obtain the fundamental frequency of the interference. Second, a series of discrete-time oscillators are used to generate interference harmonics, where harmonic phase and amplitude are obtained using the recursive least squares (RLS) algorithm. Third, the estimated interference harmonics are removed without distorting the neural signals at the interference frequencies. The simple structure and adequate numerical behavior of the algorithm renders it suitable for realtime implementation. Extensive experiments based on both invivo and synthesized data have been performed, where a reliable performance has been observed.