Towards Real-Time, Continuous Decoding of Gripping Force From Deep Brain Local Field Potentials

Invasive neurophysiology and whole brain connectomics for neural decoding in patients with brain implants

Brain computer interfaces (BCI) provide unprecedented spatiotemporal precision that will enable significant expansion in how numerous brain disorders are treated. Decoding dynamic patient states from brain signals with machine learning is required to leverage this precision, but a standardized framework for identifying and advancing novel clinical BCI approaches does not exist. Here, we developed a platform that integrates brain signal decoding with connectomics and demonstrate its utility across 123 hours of invasively recorded brain data from 73 neurosurgical patients treated for movement disorders, depression and epilepsy. First, we introduce connectomics-informed movement decoders that generalize across cohorts with Parkinson's disease and epilepsy from the US, Europe and China. Next, we reveal network targets for emotion decoding in left prefrontal and cingulate circuits in DBS patients with major depression. Finally, we showcase opportunities to improve seizure detection in responsive neurostimulation for epilepsy. Our platform provides rapid, high-accuracy decoding for precision medicine approaches that can dynamically adapt neuromodulation therapies in response to the individual needs of patients.

Robust Removal of Slow Artifactual Dynamics Induced by Deep Brain Stimulation in Local Field Potential Recordings Using SVD-Based Adaptive Filtering

Deep brain stimulation (DBS) is widely used as a treatment option for patients with movement disorders. In addition to its clinical impact, DBS has been utilized in the field of cognitive neuroscience, wherein the answers to several fundamental questions underpinning the mechanisms of neuromodulation in decision making rely on the ways in which a burst of DBS pulses, usually delivered at a clinical frequency, i.e., 130 Hz, perturb participants' choices. It was observed that neural activities recorded during DBS were contaminated with large artifacts, which lasts for a few milliseconds, as well as a low-frequency (slow) signal (~1-2 Hz) that can persist for hundreds of milliseconds. While the focus of most of methods for removing DBS artifacts was on the former, the artifact removal capabilities of the slow signal have not been addressed. In this work, we propose a new method based on combining singular value decomposition (SVD) and normalized adaptive filtering to remove both large (fast) and slow artifacts in local field potentials, recorded during a cognitive task in which bursts of DBS were utilized. Using synthetic data, we show that our proposed algorithm outperforms four commonly used techniques in the literature, namely, (1) normalized least mean square adaptive filtering, (2) optimal FIR Wiener filtering, (3) Gaussian model matching, and (4) moving average. The algorithm's capabilities are further demonstrated by its ability to effectively remove DBS artifacts in local field potentials recorded from the subthalamic nucleus during a verbal Stroop task, highlighting its utility in real-world applications.

The potential of convolutional neural networks for identifying neural states based on electrophysiological signals: experiments on synthetic and real patient data

Processing incoming neural oscillatory signals in real-time and decoding from them relevant behavioral or pathological states is often required for adaptive Deep Brain Stimulation (aDBS) and other brain-computer interface (BCI) applications. Most current approaches rely on first extracting a set of predefined features, such as the power in canonical frequency bands or various time-domain features, and then training machine learning systems that use those predefined features as inputs and infer what the underlying brain state is at each given time point. However, whether this algorithmic approach is best suited to extract all available information contained within the neural waveforms remains an open question. Here, we aim to explore different algorithmic approaches in terms of their potential to yield improvements in decoding performance based on neural activity such as measured through local field potentials (LFPs) recordings or electroencephalography (EEG). In particular, we aim to explore the potential of end-to-end convolutional neural networks, and compare this approach with other machine learning methods that are based on extracting predefined feature sets. To this end, we implement and train a number of machine learning models, based either on manually constructed features or, in the case of deep learning-based models, on features directly learnt from the data. We benchmark these models on the task of identifying neural states using simulated data, which incorporates waveform features previously linked to physiological and pathological functions. We then assess the performance of these models in decoding movements based on local field potentials recorded from the motor thalamus of patients with essential tremor. Our findings, derived from both simulated and real patient data, suggest that end-to-end deep learning-based methods may surpass feature-based approaches, particularly when the relevant patterns within the waveform data are either unknown, difficult to quantify, or when there may be, from the point of view of the predefined feature extraction pipeline, unidentified features that could contribute to decoding performance. The methodologies proposed in this study might hold potential for application in adaptive deep brain stimulation (aDBS) and other brain-computer interface systems.

Hyper-parameter tuning and feature extraction for asynchronous action detection from sub-thalamic nucleus local field potentials

Introduction

Decoding brain states from subcortical local field potentials (LFPs) indicative of activities such as voluntary movement, tremor, or sleep stages, holds significant potential in treating neurodegenerative disorders and offers new paradigms in brain-computer interface (BCI). Identified states can serve as control signals in coupled human-machine systems, e.g., to regulate deep brain stimulation (DBS) therapy or control prosthetic limbs. However, the behavior, performance, and efficiency of LFP decoders depend on an array of design and calibration settings encapsulated into a single set of hyper-parameters. Although methods exist to tune hyper-parameters automatically, decoders are typically found through exhaustive trial-and-error, manual search, and intuitive experience.

Methods

This study introduces a Bayesian optimization (BO) approach to hyper-parameter tuning, applicable through feature extraction, channel selection, classification, and stage transition stages of the entire decoding pipeline. The optimization method is compared with five real-time feature extraction methods paired with four classifiers to decode voluntary movement asynchronously based on LFPs recorded with DBS electrodes implanted in the subthalamic nucleus of Parkinson's disease patients.

Results

Detection performance, measured as the geometric mean between classifier specificity and sensitivity, is automatically optimized. BO demonstrates improved decoding performance from initial parameter setting across all methods. The best decoders achieve a maximum performance of 0.74 ± 0.06 (mean ± SD across all participants) sensitivity-specificity geometric mean. In addition, parameter relevance is determined using the BO surrogate models.

Discussion

Hyper-parameters tend to be sub-optimally fixed across different users rather than individually adjusted or even specifically set for a decoding task. The relevance of each parameter to the optimization problem and comparisons between algorithms can also be difficult to track with the evolution of the decoding problem. We believe that the proposed decoding pipeline and BO approach is a promising solution to such challenges surrounding hyper-parameter tuning and that the study's findings can inform future design iterations of neural decoders for adaptive DBS and BCI.

Movement decoding using spatio-spectral features of cortical and subcortical local field potentials

The first commercially sensing enabled deep brain stimulation (DBS) devices for the treatment of movement disorders have recently become available. In the future, such devices could leverage machine learning based brain signal decoding strategies to individualize and adapt therapy in real-time. As multi-channel recordings become available, spatial information may provide an additional advantage for informing machine learning models. To investigate this concept, we compared decoding performances from single channels vs. spatial filtering techniques using intracerebral multitarget electrophysiology in Parkinson's disease patients undergoing DBS implantation. We investigated the feasibility of spatial filtering in invasive neurophysiology and the putative utility of combined cortical ECoG and subthalamic local field potential signals for decoding grip-force, a well-defined and continuous motor readout. We found that adding spatial information to the model can improve decoding (6% gain in decoding), but the spatial patterns and additional benefit was highly individual. Beyond decoding performance results, spatial filters and patterns can be used to obtain meaningful neurophysiological information about the brain networks involved in target behavior. Our results highlight the importance of individualized approaches for brain signal decoding, for which multielectrode recordings and spatial filtering can improve precision medicine approaches for clinical brain computer interfaces.

Electrocorticography is superior to subthalamic local field potentials for movement decoding in Parkinson's disease

Brain signal decoding promises significant advances in the development of clinical brain computer interfaces (BCI). In Parkinson's disease (PD), first bidirectional BCI implants for adaptive deep brain stimulation (DBS) are now available. Brain signal decoding can extend the clinical utility of adaptive DBS but the impact of neural source, computational methods and PD pathophysiology on decoding performance are unknown. This represents an unmet need for the development of future neurotechnology. To address this, we developed an invasive brain-signal decoding approach based on intraoperative sensorimotor electrocorticography (ECoG) and subthalamic LFP to predict grip-force, a representative movement decoding application, in 11 PD patients undergoing DBS. We demonstrate that ECoG is superior to subthalamic LFP for accurate grip-force decoding. Gradient boosted decision trees (XGBOOST) outperformed other model architectures. ECoG based decoding performance negatively correlated with motor impairment, which could be attributed to subthalamic beta bursts in the motor preparation and movement period. This highlights the impact of PD pathophysiology on the neural capacity to encode movement vigor. Finally, we developed a connectomic analysis that could predict grip-force decoding performance of individual ECoG channels across patients by using their connectomic fingerprints. Our study provides a neurophysiological and computational framework for invasive brain signal decoding to aid the development of an individualized precision-medicine approach to intelligent adaptive DBS.

Machine learning based brain signal decoding for intelligent adaptive deep brain stimulation

Sensing enabled implantable devices and next-generation neurotechnology allow real-time adjustments of invasive neuromodulation. The identification of symptom and disease-specific biomarkers in invasive brain signal recordings has inspired the idea of demand dependent adaptive deep brain stimulation (aDBS). Expanding the clinical utility of aDBS with machine learning may hold the potential for the next breakthrough in the therapeutic success of clinical brain computer interfaces. To this end, sophisticated machine learning algorithms optimized for decoding of brain states from neural time-series must be developed. To support this venture, this review summarizes the current state of machine learning studies for invasive neurophysiology. After a brief introduction to the machine learning terminology, the transformation of brain recordings into meaningful features for decoding of symptoms and behavior is described. Commonly used machine learning models are explained and analyzed from the perspective of utility for aDBS. This is followed by a critical review on good practices for training and testing to ensure conceptual and practical generalizability for real-time adaptation in clinical settings. Finally, first studies combining machine learning with aDBS are highlighted. This review takes a glimpse into the promising future of intelligent adaptive DBS (iDBS) and concludes by identifying four key ingredients on the road for successful clinical adoption: i) multidisciplinary research teams, ii) publicly available datasets, iii) open-source algorithmic solutions and iv) strong world-wide research collaborations.

Continuously Decoding Grasping Movements using Stereotactic Depth Electrodes

Brain-Computer Interfaces (BCIs) that decode a patient's movement intention to control a prosthetic device could restore some independence to paralyzed patients. An important step on the road towards naturalistic prosthetic control is to decode movement continuously with low-latency. BCIs based on intracortical micro-arrays provide continuous control of robotic arms, but require a minor craniotomy. Surface recordings of neural activity using EEG have made great advances over the last years, but suffer from high noise levels and large intra-session variance. Here, we investigate the use of minimally invasive recordings using stereotactically implanted EEG (sEEG). These electrodes provide a sparse sampling across many brain regions. So far, promising decoding results have been presented using data measured from the subthalamic nucleus or trial-to-trial based methods using depth electrodes. In this work, we demonstrate that grasping movements can continuously be decoded using sEEG electrodes, as well. Beta and high-gamma activity was extracted from eight participants performing a grasping task. We demonstrate above chance level decoding of movement vs rest and left vs right, from both frequency bands with accuracies up to 0.94 AUC. The vastly different electrode locations between participants lead to large variability. In the future, we hope that sEEG recordings will provide additional information for the decoding process in neuroprostheses.

Optimizing Time-Frequency Feature Extraction and Channel Selection through Gradient Backpropagation to Improve Action Decoding based on Subthalamic Local Field Potentials

Neural oscillating patterns, or time-frequency features, predicting voluntary motor intention, can be extracted from the local field potentials (LFPs) recorded from the sub-thalamic nucleus (STN) or thalamus of human patients implanted with deep brain stimulation (DBS) electrodes for the treatment of movement disorders. This paper investigates the optimization of signal conditioning processes using deep learning to augment time-frequency feature extraction from LFP signals, with the aim of improving the performance of real-time decoding of voluntary motor states. A brain-computer interface (BCI) pipeline capable of continuously classifying discrete pinch grip states from LFPs was designed in Pytorch, a deep learning framework. The pipeline was implemented offline on LFPs recorded from 5 different patients bilaterally implanted with DBS electrodes. Optimizing channel combination in different frequency bands and frequency domain feature extraction demonstrated improved classification accuracy of pinch grip detection and laterality of the pinch (either pinch of the left hand or pinch of the right hand). Overall, the optimized BCI pipeline achieved a maximal average classification accuracy of 79.67±10.02% when detecting all pinches and 67.06±10.14% when considering the laterality of the pinch.

Spatial-Frequency Feature Learning and Classification of Motor Imagery EEG Based on Deep Convolution Neural Network

EEG pattern recognition is an important part of motor imagery- (MI-) based brain computer interface (BCI) system. Traditional EEG pattern recognition algorithm usually includes two steps, namely, feature extraction and feature classification. In feature extraction, common spatial pattern (CSP) is one of the most frequently used algorithms. However, in order to extract the optimal CSP features, prior knowledge and complex parameter adjustment are often required. Convolutional neural network (CNN) is one of the most popular deep learning models at present. Within CNN, feature learning and pattern classification are carried out simultaneously during the procedure of iterative updating of network parameters; thus, it can remove the complicated manual feature engineering. In this paper, we propose a novel deep learning methodology which can be used for spatial-frequency feature learning and classification of motor imagery EEG. Specifically, a multilayer CNN model is designed according to the spatial-frequency characteristics of MI EEG signals. An experimental study is carried out on two MI EEG datasets (BCI competition III dataset IVa and a self-collected right index finger MI dataset) to validate the effectiveness of our algorithm in comparison with several closely related competing methods. Superior classification performance indicates that our proposed method is a promising pattern recognition algorithm for MI-based BCI system.

Electrophysiological differences between upper and lower limb movements in the human subthalamic nucleus

OBJECTIVE

Functional processes in the brain are segregated in both the spatial and spectral domain. Motivated by findings reported at the cortical level in healthy participants we test the hypothesis in the basal ganglia of Parkinson's disease patients that lower frequency beta band activity relates to motor circuits associated with the upper limb and higher beta frequencies with lower limb movements.

METHODS

We recorded local field potentials (LFPs) from the subthalamic nucleus using segmented "directional" DBS leads, during which patients performed repetitive upper and lower limb movements. Movement-related spectral changes in the beta and gamma frequency-ranges and their spatial distributions were compared between limbs.

RESULTS

We found that the beta desynchronization during leg movements is characterised by a strikingly greater involvement of higher beta frequencies (24-31 Hz), regardless of whether this was contralateral or ipsilateral to the limb moved. The spatial distribution of limb-specific movement-related changes was evident at higher gamma frequencies.

CONCLUSION

Limb processing in the basal ganglia is differentially organised in the spectral and spatial domain and can be captured by directional DBS leads.

SIGNIFICANCE

These findings may help to refine the use of the subthalamic LFPs as a control signal for adaptive DBS and neuroprosthetic devices.