Deep learning models have demonstrated exceptional performance in a variety of real-world applications. These successes are often attributed to strong base models that can generalize to novel tasks with limited supporting data while keeping prior knowledge intact. However, these impressive results are based on the availability of a large amount of high-quality data, which is often lacking in specialized biomedical applications. In such fields, models are often developed with limited data that arrive incrementally with novel categories. This requires the model to adapt to new information while preserving existing knowledge. Few-Shot Class-Incremental Learning (FSCIL) methods offer a promising approach to addressing these challenges, but they also depend on strong base models that face the same aforementioned limitations. To overcome these constraints, we propose AnchorInv following the straightforward and efficient buffer-replay strategy. Instead of selecting and storing raw data, AnchorInv generates synthetic samples guided by anchor points in the feature space. This approach protects privacy and regularizes the model for adaptation. When evaluated on three public physiological time series datasets, AnchorInv exhibits efficient knowledge forgetting prevention and improved adaptation to novel classes, surpassing state-of-the-art baselines.

Existing neurostimulation systems implanted for the treatment of neurodegenerative disorders generally deliver invariable therapy parameters, regardless of phase of the sleep/wake cycle. However, there is considerable evidence that brain activity in these conditions varies according to this cycle, with discrete patterns of dysfunction linked to loss of circadian rhythmicity, worse clinical outcomes and impaired patient quality of life. We present a targeted concept of circadian neuromodulation using a novel device platform. This system utilises stimulation of circuits important in sleep and wake regulation, delivering bioelectronic cues (Zeitgebers) aimed at entraining rhythms to more physiological patterns in a personalised and fully configurable manner. Preliminary evidence from its first use in a clinical trial setting, with brainstem arousal circuits as a surgical target, further supports its promising impact on sleep/wake pathology. Data included in this paper highlight its versatility and effectiveness on two different patient phenotypes. In addition to exploring acute and long-term electrophysiological and behavioural effects, we also discuss current caveats and future feature improvements of our proposed system, as well as its potential applicability in modifying disease progression in future therapies.

Sleep Stage Classification (SSC) is a labor-intensive task, requiring experts to examine hours of electrophysiological recordings for manual classification. This is a limiting factor when it comes to leveraging sleep stages for therapeutic purposes. With increasing affordability and expansion of wearable devices, automating SSC may enable deployment of sleep-based therapies at scale. Deep Learning has gained increasing attention as a potential method to automate this process. Previous research has shown accuracy comparable to manual expert scores. However, previous approaches require sizable amount of memory and computational resources. This constrains the ability to classify in real time and deploy models on the edge. To address this gap, we aim to provide a model capable of predicting sleep stages in real-time, without requiring access to external computational sources (e.g., mobile phone, cloud). The algorithm is power efficient to enable use on embedded battery powered systems. Our compact sleep stage classifier can be deployed on most off-the-shelf microcontrollers (MCU) with constrained hardware settings. This is due to the memory footprint of our approach requiring significantly fewer operations. The model was tested on three publicly available data bases and achieved performance comparable to the state of the art, whilst reducing model complexity by orders of magnitude (up to 280 times smaller compared to state of the art). We further optimized the model with quantization of parameters to 8 bits with only an average drop of 0.95% in accuracy. When implemented in firmware, the quantized model achieves a latency of 1.6 seconds on an Arm Cortex-M4 processor, allowing its use for on-line SSC-based therapies.

Stimulation optimization has garnered considerable interest in recent years in order to efficiently parametrize neuromodulation-based therapies. To date, efforts focused on automatically identifying settings from parameter spaces that do not change over time. A limitation of these approaches, however, is that they lack consideration for time dependent factors that may influence therapy outcomes. Disease progression and biological rhythmicity are two sources of variation that may influence optimal stimulation settings over time. To account for this, we present a novel time-varying Bayesian optimization (TV-BayesOpt) for tracking the optimum parameter set for neuromodulation therapy. We evaluate the performance of TV-BayesOpt for tracking gradual and periodic slow variations over time. The algorithm was investigated within the context of a computational model of phase-locked deep brain stimulation for treating oscillopathies representative of common movement disorders such as Parkinson's disease and Essential Tremor. When the optimal stimulation settings changed due to gradual and periodic sources, TV-BayesOpt outperformed standard time-invariant techniques and was able to identify the appropriate stimulation setting. Through incorporation of both a gradual "forgetting" and periodic covariance functions, the algorithm maintained robust performance when a priori knowledge differed from observed variations. This algorithm presents a broad framework that can be leveraged for the treatment of a range of neurological and psychiatric conditions and can be used to track variations in optimal stimulation settings such as amplitude, pulse-width, frequency and phase for invasive and non-invasive neuromodulation strategies.

Electrical stimulation of deep brain structures is an established therapy for drug-resistant focal epilepsy. The emerging implantable neural sensing and stimulating (INSS) technology enables simultaneous delivery of chronic deep brain stimulation (DBS) and recording of electrical brain activity from deep brain structures while patients live in their home environment. Long-term intracranial electroencephalography (iEEG) iEEG signals recorded by INSS devices represent an opportunity to investigate brain neurophysiology and how DBS affects neural circuits. However, novel algorithms and data processing pipelines need to be developed to facilitate research of these long-term iEEG signals. Early-stage analytical infrastructure development for INSS applications can be limited by lacking iEEG data that might not always be available. Here, we investigate the feasibility of utilizing the Deep Generative Adversarial Network (DCGAN) for synthetic iEEG data generation. We trained DCGAN using 3-second iEEG segments and validated synthetic iEEG usability by training a classification model, using synthetic iEEG only and providing a good classification performance on unseen real iEEG with an F1 score 0.849. Subsequently, we demonstrated the feasibility of utilizing the synthetic iEEG in the INSS application development by training a deep learning network for DBS artifact removal using synthetic data only and demonstrated the performance on real iEEG signals. The presented strategy of on-demand generating synthetic iEEG will benefit early-stage algorithm development for INSS applications.

Recent development in implantable devices for electrical brain stimulation includes sensing and embedded computing capabilities that enable adaptive stimulation strategies. Applications include stimulation triggered by pathologic brain activity and endogenous rhythms, such as circadian rhythms. We developed and tested a system that integrates an electrical brain stimulation & sensing implantable device with embedded computing and uses a distributed system with commercial electronics, smartphone and smartwatch for patient annotations, extensive behavioral testing, and adaptive stimulation in subjects in their natural environments. The system enables precise time synchronization of the external components with the brain stimulating device and is coupled with automated analysis of continuous streaming electrophysiology synchronized with patient reports. The system leverages a real-time bi-directional interface between devices and patients with epilepsy living in their natural environment.

Brain stimulation therapies have been established as effective treatments for Parkinson's disease, essential tremor, and epilepsy, as well as having high diagnostic and therapeutic potential in a wide range of neurological and psychiatric conditions. Novel interventions such as extended reality (XR), video games and exergames that can improve physiological and cognitive functioning are also emerging as targets for therapeutic and rehabilitative treatments. Previous studies have proposed specific applications involving non-invasive brain stimulation (NIBS) and virtual environments, but to date these have been uni-directional and restricted to specific applications or proprietary hardware. Here, we describe technology integration methods that enable invasive and non-invasive brain stimulation devices to interface with a cross-platform game engine and development platform for creating bi-directional brain-computer interfaces (BCI) and XR-based interventions. Furthermore, we present a highly-modifiable software framework and methods for integrating deep brain stimulation (DBS) in 2D, 3D, virtual and mixed reality applications, as well as extensible applications for BCI integration in wireless systems. The source code and integrated brain stimulation applications are available online at https://github.com/oxfordbioelectronics/brain-stim-game.

The accurate measurement of brain activity by Brain-Machine-Interfaces (BMI) and closed-loop Deep Brain Stimulators (DBS) is one of the most important steps in communicating between the brain and subsequent processing blocks. In conventional chest-mounted systems, frequently used in DBS, a significant amount of artifact can be induced in the sensing interface, often as a common-mode signal applied between the case and the sensing electrodes. Attenuating this common-mode signal can be a serious challenge in these systems due to finite common-mode-rejection-ratio (CMRR) capability in the interface. Emerging BMI and DBS devices are being developed which can mount on the skull. Mounting the system on the cranial region can potentially suppress these induced physiological signals by limiting the artifact amplitude. In this study, we model the effect of artifacts by focusing on cardiac activity, using a current- source dipole model in a torso-shaped volume conductor. Performing finite element simulation with the different DBS architectures, we estimate the ECG common mode artifacts for several device architectures. Using this model helps define the overall requirements for the total system CMRR to maintain resolution of brain activity. The results of the simulations estimate that the cardiac artifacts for skull-mounted systems will have a significantly lower effect than non-cranial systems that include the pectoral region. It is expected that with a pectoral mounted device, a minimum of 60-80 dB CMRR is required to suppress the ECG artifact, depending on device placement relative to the cardiac dipole, while in cranially mounted devices, a 0 dB CMRR is sufficient, in the worst-case scenario. In addition, the model suggests existing commercial devices could optimize performance with a right-hand side placement. The methods used for estimating cardiac artifacts can be extended to other sources such as motion/muscle sources. The susceptibility of the device to artifacts has significant implications for the practical translation of closed-loop DBS and BMI, including the choice of biomarkers, the system design requirements, and the surgical placement of the device relative to artifact sources.

Deep brain stimulation (DBS) for Parkinson's disease, essential tremor and epilepsy is an established palliative treatment. DBS uses electrical neuromodulation to suppress symptoms. Most current systems provide a continuous pattern of fixed stimulation, with clinical follow-ups to refine settings constrained to normal office hours. An issue with this management strategy is that the impact of stimulation on circadian, i.e. sleep-wake, rhythms is not fully considered; either in the device design or in the clinical follow-up. Since devices can be implanted in brain targets that couple into the reticular activating network, impact on wakefulness and sleep can be significant. This issue will likely grow as new targets are explored, with the potential to create entraining signals that are uncoupled from environmental influences. To address this issue, we have designed a new brain-machine-interface for DBS that combines a slow-adaptive circadian-based stimulation pattern with a fast-acting pathway for responsive stimulation, demonstrated here for seizure management. In preparation for first-in-human research trials to explore the utility of multi-timescale automated adaptive algorithms, design and prototyping was carried out in line with ISO risk management standards, ensuring patient safety. The ultimate aim is to account for chronobiology within the algorithms embedded in brain-machine-interfaces and in neuromodulation technology more broadly.

Millions of people worldwide suffer from neurological disorders such as epilepsy, movement disorders, and obsessive-compulsive disorder (OCD), depression, and delirium. To provide relief from these disorders, brain stimulation therapies have been shown to be effective at controlling onsets of seizures, tremors, dyskinesia, dystonia, and OCD episodes. Current development of brain stimulation therapies has pivoted toward closed-loop control of sensing onset events and correspondingly delivering adaptive stimulation. Development of closed-loop brain stimulation therapies for neurological disorders rely on the identification of neural biomarkers. As such, a brain signal monitoring system that can chronically record these neurological events is essential to the continued development of neuromodulation systems and therapies. Through analyzing clinical data, neural disorder biomarkers can be identified and novel therapies can be optimized. This paper outlines the development of a translational deep brain stimulation monitoring system utilizing Medtronic's RC+S System to help clinicians and patients accurately record and document neural disorder onset events. With this neural data, stimulation therapy parameters can be adjusted using the system without requiring an in-person office visit. The system is capable of wirelessly communicating with multiple implanted neurostimulators, monitoring disorder onset biomarkers, and periodically downloading real-time brain signal data as well as loop recordings triggered by device-detected disorder onset events. This translational system and neural disorder onset data can be used to optimize therapies, minimize symptom onsets, enable episodic care management, and improve chronic care management.

The monitoring of neuronal activity could potentially expand the diagnostic and therapeutic capabilities of neuroprosthesis. The challenge of designing sensing and control systems is two-fold: first, the signal input must be robust for chronic recording; second, the circuit architecture must be capable of achieving signal processing, algorithm control, and telemetry with a limited power budget. The first requirement should be met by measuring field potentials, which represent ensemble behavior in a neural network and can be measured chronically. For the second requirement, architecting an effective solution requires identification of the key information of interest and partitioning the signal chain to play to the strengths of analog vs. digital processing. For many neurological states of interest, information 'biomarkers' are encoded as low frequency power fluctuations within well-defined frequency bands of field potentials, similar to the amplitude modulation found in an AM radio. Recognizing this similarity, the feasibility prototype adapts a chopper-stabilized instrumentation amplifier to act as a superheterodyning AM receiver for brain signals. Since the physiological power fluctuations are generally orders of magnitude slower than the frequency at which they are encoded, the use of efficient analog preprocessing greatly reduces the overall energy requirements for implementing a complete mixed- signal system. Since the science of field potentials is rapidly evolving, the superheterodyning chopper is advantageous given its flexibility and immunity to process, temperature, and mismatch variations. This paper will discuss the design of a complete system prototype for a neurostimulator research tool; the design has a noise floor of under 2muVrms and a total system current of 25muW/processing channel (1.8V supply) while performing biomarker extraction, algorithmic processing and control, and data loop recording.