Machine learning seizure prediction: one problematic but accepted practice.

Objective.Epilepsy is one of the most common neurological disorders and can have a devastating effect on a person's quality of life. As such, the search for markers which indicate an upcoming seizure is a critically important area of research which would allow either on-demand treatment or early warning for people suffering with these disorders. There is a growing body of work which uses machine learning methods to detect pre-seizure biomarkers from electroencephalography (EEG), however the high prediction rates published do not translate into the clinical setting. Our objective is to investigate a potential reason for this.Approach.We conduct an empirical study of a commonly used data labelling method for EEG seizure prediction which relies on labelling small windows of EEG data in temporal groups then selecting randomly from those windows to validate results. We investigate a confound for this approach for seizure prediction and demonstrate the ease at which it can be inadvertently learned by a machine learning system.Main results.We find that non-seizure signals can create decision surfaces for machine learning approaches which can result in false high prediction accuracy on validation datasets. We prove this by training an artificial neural network to learn fake seizures (fully decoupled from biology) in real EEG.Significance.The significance of our findings is that many existing works may be reporting results based on this confound and that future work should adhere to stricter requirements in mitigating this confound. The problematic, but commonly accepted approach in the literature for seizure prediction labelling is potentially preventing real advances in developing solutions for these sufferers. By adhering to the guidelines in this paper future work in machine learning seizure prediction is more likely to be clinically relevant.

Robustness of Beta Desynchronization from Chronically Implanted Cortical Electrodes on Multiple Time Scales.

Adaptive deep brain stimulation (aDBS) promises a significant improvement in patient outcomes, compared to existing deep brain stimulation devices. Fully implanted systems represent the next step to the clinical adoption of aDBS. We take advantage of a unique longitudinal data set formed as part of an effort to investigate aDBS for essential tremor to verify the long term reliability of electrocorticography strips over the motor cortex as a source of bio-markers for control of adaptive stimulation. We show that beta band event related de-synchronization, a promising bio-marker for movement, is robust even when used to trigger aDBS. Over the course of several months we show a minor increase in beta band event related de-synchronization in patients with active deep brain stimulation confirming that it could be used in chronically implanted systems.Clinical relevance - We show the promise and practicality of cortical electrocorticography strips for use in fully implanted, clinically translatable, aDBS systems.

Closing the Loop With Cortical Sensing: The Development of Adaptive Deep Brain Stimulation for Essential Tremor Using the Activa PC+S.

Deep Brain Stimulation (DBS) is an important tool in the treatment of pharmacologically resistant neurological movement disorders such as essential tremor (ET) and Parkinson's disease (PD). However, the open-loop design of current systems may be holding back the true potential of invasive neuromodulation. In the last decade we have seen an explosion of activity in the use of feedback to "close the loop" on neuromodulation in the form of adaptive DBS (aDBS) systems that can respond to the patient's therapeutic needs. In this paper we summarize the accomplishments of a 5-year study at the University of Washington in the use of neural feedback from an electrocorticography strip placed over the sensorimotor cortex. We document our progress from an initial proof of hardware all the way to a fully implanted adaptive stimulation system that leverages machine-learning approaches to simplify the programming process. In certain cases, our systems out-performed current open-loop approaches in both power consumption and symptom suppression. Throughout this effort, we collaborated with neuroethicists to capture patient experiences and take them into account whilst developing ethical aDBS approaches. Based on our results we identify several key areas for future work. "Graded" aDBS will allow the system to smoothly tune the stimulation level to symptom severity, and frequent automatic calibration of the algorithm will allow aDBS to adapt to the time-varying dynamics of the disease without additional input from a clinician. Additionally, robust computational models of the pathophysiology of ET will allow stimulation to be optimized to the nuances of an individual patient's symptoms. We also outline the unique advantages of using cortical electrodes for control and the remaining hardware limitations that need to be overcome to facilitate further development in this field. Over the course of this study we have verified the potential of fully-implanted, cortically driven aDBS as a feasibly translatable treatment for pharmacologically resistant ET.

A Pilot Study on Data-Driven Adaptive Deep Brain Stimulation in Chronically Implanted Essential Tremor Patients.

Deep brain stimulation (DBS) is an established therapy for Parkinson's disease (PD) and essential-tremor (ET). In adaptive DBS (aDBS) systems, online tuning of stimulation parameters as a function of neural signals may improve treatment efficacy and reduce side-effects. State-of-the-art aDBS systems use symptom surrogates derived from neural signals-so-called neural markers (NMs)-defined on the patient-group level, and control strategies assuming stationarity of symptoms and NMs. We aim at improving these aDBS systems with (1) a data-driven approach for identifying patient- and session-specific NMs and (2) a control strategy coping with short-term non-stationary dynamics. The two building blocks are implemented as follows: (1) The data-driven NMs are based on a machine learning model estimating tremor intensity from electrocorticographic signals. (2) The control strategy accounts for local variability of tremor statistics. Our study with three chronically implanted ET patients amounted to five online sessions. Tremor quantified from accelerometer data shows that symptom suppression is at least equivalent to that of a continuous DBS strategy in 3 out-of 4 online tests, while considerably reducing net stimulation (at least 24%). In the remaining online test, symptom suppression was not significantly different from either the continuous strategy or the no treatment condition. We introduce a novel aDBS system for ET. It is the first aDBS system based on (1) a machine learning model to identify session-specific NMs, and (2) a control strategy coping with short-term non-stationary dynamics. We show the suitability of our aDBS approach for ET, which opens the door to its further study in a larger patient population.

Rebound effect in deep brain stimulation for essential tremor and symptom severity estimation from neural data.

Deep brain stimulation (DBS) is a safe and established treatment for essential tremor (ET). However, there remains considerable room for improvement due to concerns associated with the initial implant surgery, semi-regular revision surgeries for battery replacements, and side effects including paresthesia, gait ataxia, and emotional disinhibition that have been associated with continuous, or conventional, DBS (cDBS) treatment. Adaptive DBS (aDBS) seeks to ameliorate some of these concerns by using feedback from either an external wearable or implanted sensor to modulate stimulation parameters as needed. aDBS has been demonstrated to be as or more effective than cDBS, but the purely binary control system most commonly deployed by aDBS systems likely still provides sub-optimal treatment and may introduce new issues. One example of these issues is rebound effect, in which the tremor symptoms of an ET patient receiving DBS therapy temporarily worsen after cessation of stimulation before leveling out to a steady state. Here is presented a quantitative analysis of rebound effect in 3 patients receiving DBS for ET. Rebound was evident in all 3 patients by both clinical assessment and inertial measurement unit data, peaking by the latter at Tp = 6.65 minutes after cessation of stimulation. Using features extracted from neural data, linear regression was applied to predict tremor severity, with $R_{avg{\text{ }}}^2 = 0.82$. These results strongly suggest that rebound effect and the additional information made available by rebound effect should be considered and exploited when designing novel aDBS systems.

A tablet- and mobile-based application for remote diagnosis and analysis of movement disorder symptoms.

One significant hindrance to effective diagnosis of movement disorders (MDs) and analysis of their progression is the requirement for patients to conduct tests in the presence of a clinician. Here is presented a pilot study for diagnosis of essential tremor (ET), the world's most common MD, through analysis of a tablet- or mobile-based drawing task that may be selected at will, with the spiral- and line-drawing tasks of the Fahn-Tolosa-Marin tremor rating scale serving as our task in this work. This system replaces the need for pen-and-paper drawing tests while permitting advanced quantitative analysis of drawing smoothness, pressure applied, and other measures. Data is securely recorded and stored in the cloud, from which all analysis was conducted remotely. This will enable longitudinal analysis of patient disease progression without the need for excessive clinical visits. Several features were extracted and recursive feature elimination applied to rank the features' individual contribution to our classifier. Maximum cross-validated classification accuracy on a preliminary sample set was 98.3%. Future work will involve collecting healthy subject data from an age-controlled population and extending this diagnostic application to additional conditions, as well as incorporating regression-based symptom severity analysis. This highly promising new technology has the potential to substantially alleviate the demands placed on both clinicians and patients by bringing MD treatment more into line with the era of personalized medicine.

Automated Deep Brain Stimulation Programming for Tremor.

Deep brain stimulation (DBS) programming, the systematic selection of fixed electrical stimulation parameters that deliver maximal therapeutic benefit while limiting side effects, poses several challenges in the treatment of movement disorders. DBS programming requires the expertise of trained neurologists or nurses who assess patient symptoms according to standardized clinical rating scales and use patient reports of DBS-related side effects to adjust stimulation parameters and optimize therapy. In this paper, we describe and validate an automated software platform for DBS programming for tremor associated with Parkinson's disease and essential tremor. DBS parameters are changed automatically through a direct computer interface with implanted neurostimulators. Each tested DBS setting is ranked according to its effect on tremor, which is assessed using smartwatch inertial measurement unit data, and side effects, which are reported through a user interface. Blinded neurologist assessments showed the automated programming method performed at least as well as clinician mediated programming in selecting the optimal settings for tremor therapy. This proof of concept study describes a novel DBS programming paradigm that may improve programming efficiency and outcomes, increase access to programming outside specialty clinics, and aid in the development of adaptive and closed-loop DBS strategies.

Adaptive deep brain stimulation for Parkinson's disease using motor cortex sensing.

OBJECTIVE: Contemporary deep brain stimulation (DBS) for Parkinson's disease is delivered continuously, and adjustments based on patient's changing symptoms must be made manually by a trained clinician. Patients may be subjected to energy intensive settings at times when they are not needed, possibly resulting in stimulation-induced adverse effects, such as dyskinesia. One solution is 'adaptive' DBS, in which stimulation is modified in real time based on neural signals that co-vary with the severity of motor signs or of stimulation-induced adverse effects. Here we show the feasibility of adaptive DBS using a fully implanted neural prosthesis. APPROACH: We demonstrate adaptive deep brain stimulation in two patients with Parkinson's disease using a fully implanted neural prosthesis that is enabled to utilize brain sensing to control stimulation amplitude (Activa PC + S). We used a cortical narrowband gamma (60-90 Hz) oscillation related to dyskinesia to decrease stimulation voltage when gamma oscillatory activity is high (indicating dyskinesia) and increase stimulation voltage when it is low. MAIN RESULTS: We demonstrate the feasibility of 'adaptive deep brain stimulation' in two patients with Parkinson's disease. In short term in-clinic testing, energy savings were substantial (38%-45%), and therapeutic efficacy was maintained. SIGNIFICANCE: This is the first demonstration of adaptive DBS in Parkinson's disease using a fully implanted device and neural sensing. Our approach is distinct from other strategies utilizing basal ganglia signals for feedback control.

Run-to-Run Optimization Control Within Exact Inverse Framework for Scan Tracking.

A run-to-run optimization controller uses a reduced set of measurement parameters, in comparison to more general feedback controllers, to converge to the best control point for a repetitive process. A new run-to-run optimization controller is presented for the scanning fiber device used for image acquisition and display. This controller utilizes very sparse measurements to estimate a system energy measure and updates the input parameterizations iteratively within a feedforward with exact-inversion framework. Analysis, simulation, and experimental investigations on the scanning fiber device demonstrate improved scan accuracy over previous methods and automatic controller adaptation to changing operating temperature. A specific application example and quantitative error analyses are provided of a scanning fiber endoscope that maintains high image quality continuously across a 20 °C temperature rise without interruption of the 56 Hz video.

Chronic electrocorticography for sensing movement intention and closed-loop deep brain stimulation with wearable sensors in an essential tremor patient.

Deep brain stimulation (DBS) has become a widespread and valuable treatment for patients with movement disorders such as essential tremor (ET). However, current DBS treatment constantly delivers stimulation in an open loop, which can be inefficient. Closing the loop with sensors to provide feedback may increase power efficiency and reduce side effects for patients. New implantable neuromodulation platforms, such as the Medtronic Activa PC+S DBS system, offer important data sources by providing chronic neural sensing capabilities and a means of investigating dynamic stimulation based on symptom measurements. The authors implanted in a single patient with ET an Activa PC+S system, a cortical strip of electrodes on the hand sensorimotor cortex, and therapeutic electrodes in the ventral intermediate nucleus of the thalamus. In this paper they describe the effectiveness of the platform when sensing cortical movement intentions while the patient actually performed and imagined performing movements. Additionally, they demonstrate dynamic closed-loop DBS based on several wearable sensor measurements of tremor intensity.

Proceedings of the Fourth Annual Deep Brain Stimulation Think Tank: A Review of Emerging Issues and Technologies.

This paper provides an overview of current progress in the technological advances and the use of deep brain stimulation (DBS) to treat neurological and neuropsychiatric disorders, as presented by participants of the Fourth Annual DBS Think Tank, which was convened in March 2016 in conjunction with the Center for Movement Disorders and Neurorestoration at the University of Florida, Gainesveille FL, USA. The Think Tank discussions first focused on policy and advocacy in DBS research and clinical practice, formation of registries, and issues involving the use of DBS in the treatment of Tourette Syndrome. Next, advances in the use of neuroimaging and electrochemical markers to enhance DBS specificity were addressed. Updates on ongoing use and developments of DBS for the treatment of Parkinson's disease, essential tremor, Alzheimer's disease, depression, post-traumatic stress disorder, obesity, addiction were presented, and progress toward innovation(s) in closed-loop applications were discussed. Each section of these proceedings provides updates and highlights of new information as presented at this year's international Think Tank, with a view toward current and near future advancement of the field.

Proceedings of the Third Annual Deep Brain Stimulation Think Tank: A Review of Emerging Issues and Technologies.

The proceedings of the 3rd Annual Deep Brain Stimulation Think Tank summarize the most contemporary clinical, electrophysiological, imaging, and computational work on DBS for the treatment of neurological and neuropsychiatric disease. Significant innovations of the past year are emphasized. The Think Tank's contributors represent a unique multidisciplinary ensemble of expert neurologists, neurosurgeons, neuropsychologists, psychiatrists, scientists, engineers, and members of industry. Presentations and discussions covered a broad range of topics, including policy and advocacy considerations for the future of DBS, connectomic approaches to DBS targeting, developments in electrophysiology and related strides toward responsive DBS systems, and recent developments in sensor and device technologies.

Motor cortical decoding performance depends on controlled system order.

Recent advances in intracortical brain-machine interfaces (BMIs) for position control have leveraged state estimators to decode intended movements from cortical activity. We revisit the underlying assumptions behind the use of Kalman filters in this context, focusing on the fact that identified cortical coding models capture closed-loop task dynamics. We show that closed-loop models can be partitioned, exposing feedback policies of the brain which are separate from interface and task dynamics. Changing task dynamics may cause the brain to change its control policy, and consequently the closed-loop dynamics. This may degrade performance of decoders upon switching from manual tasks to velocity-controlled BMI-mediated tasks. We provide experimental results showing that for the same manual cursor task, changing system order affects neural coding of movement. In one experimental condition force determines position directly, and in the other force determines cursor velocity. From this we draw an analogy to subjects transitioning from manual reaching tasks to velocity-controlled BMI tasks. We conclude with suggested principles for improving BMI decoder performance, including matching the controlled system order between manual and brain control, and identifying the brain's controller dynamics rather than complete closed-loop dynamics.

Multiscale modeling of cardiac cellular energetics.

Multiscale modeling is essential to integrating knowledge of human physiology starting from genomics, molecular biology, and the environment through the levels of cells, tissues, and organs all the way to integrated systems behavior. The lowest levels concern biophysical and biochemical events. The higher levels of organization in tissues, organs, and organism are complex, representing the dynamically varying behavior of billions of cells interacting together. Models integrating cellular events into tissue and organ behavior are forced to resort to simplifications to minimize computational complexity, thus reducing the model's ability to respond correctly to dynamic changes in external conditions. Adjustments at protein and gene regulatory levels shortchange the simplified higher-level representations. Our cell primitive is composed of a set of subcellular modules, each defining an intracellular function (action potential, tricarboxylic acid cycle, oxidative phosphorylation, glycolysis, calcium cycling, contraction, etc.), composing what we call the "eternal cell," which assumes that there is neither proteolysis nor protein synthesis. Within the modules are elements describing each particular component (i.e., enzymatic reactions of assorted types, transporters, ionic channels, binding sites, etc.). Cell subregions are stirred tanks, linked by diffusional or transporter-mediated exchange. The modeling uses ordinary differential equations rather than stochastic or partial differential equations. This basic model is regarded as a primitive upon which to build models encompassing gene regulation, signaling, and long-term adaptations in structure and function. During simulation, simpler forms of the model are used, when possible, to reduce computation. However, when this results in error, the more complex and detailed modules and elements need to be employed to improve model realism. The processes of error recognition and of mapping between different levels of model form complexity are challenging but are essential for successful modeling of large-scale systems in reasonable time. Currently there is to this end no established methodology from computational sciences.