Gait is a core motor function and is impaired in numerous neurological diseases, including Parkinson’s disease (PD). Treatment changes in PD are frequently driven by gait assessments in the clinic, commonly rated as part of the Movement Disorder Society (MDS) Unified PD Rating Scale (UPDRS) assessment (item 3.10). We proposed and evaluated a novel approach for estimating severity of gait impairment in Parkinson’s disease using a computer vision-based methodology. The system we developed can be used to obtain an estimate for a rating to catch potential errors, or to gain an initial rating in the absence of a trained clinician—for example, during remote home assessments. Videos ($n=729$) were collected as part of routine MDS-UPDRS gait assessments of Parkinson’s patients, and a deep learning library was used to extract body key-point coordinates for each frame. Data were recorded at five clinical sites using commercially available mobile phones or tablets, and had an associated severity rating from a trained clinician. Six features were calculated from time-series signals of the extracted key-points. These features characterized key aspects of the movement including speed (step frequency, estimated using a novel Gamma-Poisson Bayesian model), arm swing, postural control and smoothness (or roughness) of movement. An ordinal random forest classification model (with one class for each of the possible ratings) was trained and evaluated using 10-fold cross validation. Step frequency point estimates from the Bayesian model were highly correlated with manually labelled step frequencies of 606 video clips showing patients walking towards or away from the camera (Pearson’s $r=0.80$, $p<0.001$). Our classifier achieved a balanced accuracy of 50% (chance = 25%). Estimated UPDRS ratings were within one of the clinicians’ ratings in 95% of cases. There was a significant correlation between clinician labels and model estimates (Spearman’s $\rho =0.52$, $p<0.001$). We show how the interpretability of the feature values could be used by clinicians to support their decision-making and provide insight into the model’s objective UPDRS rating estimation. The severity of gait impairment in Parkinson’s disease can be estimated using a single patient video, recorded using a consumer mobile device and within standard clinical settings; i.e., videos were recorded in various hospital hallways and offices rather than gait laboratories. This approach can support clinicians during routine assessments by providing an objective rating (or second opinion), and has the potential to be used for remote home assessments, which would allow for more frequent monitoring.

Adaptive deep brain stimulation (aDBS) is a promising new technology with increasing use in experimental trials to treat a diverse array of indications such as movement disorders (Parkinson’s disease, essential tremor), psychiatric disorders (depression, OCD), chronic pain and epilepsy. In many aDBS trials, a neural biomarker of interest is compared with a predefined threshold and stimulation amplitude is adjusted accordingly. Across indications and implant locations, potential biomarkers are greatly influenced by sleep. Successful chronic embedded adaptive detectors must incorporate a strategy to account for sleep, to avoid unwanted or unexpected algorithm behavior. Here, we show a dual algorithm design with two independent detectors, one used to track sleep state (wake/sleep) and the other used to track parkinsonian motor state (medication-induced fluctuations). Across six hemispheres (four patients) and 47 days, our detector successfully transitioned to sleep mode while patients were sleeping, and resumed motor state tracking when patients were awake. Designing “sleep aware” aDBS algorithms may prove crucial for deployment of clinically effective fully embedded aDBS algorithms.

Objective. To provide a design analysis and guidance framework for the implementation of concurrent stimulation and sensing during adaptive deep brain stimulation (aDBS) with particular emphasis on artifact mitigations. Approach. We defined a general architecture of feedback-enabled devices, identified key components in the signal chain which might result in unwanted artifacts and proposed methods that might ultimately enable improved aDBS therapies. We gathered data from research subjects chronically-implanted with an investigational aDBS system, Summit RC + S, to characterize and explore artifact mitigations arising from concurrent stimulation and sensing. We then used a prototype investigational implantable device, DyNeuMo, and a bench-setup that accounts for tissue–electrode properties, to confirm our observations and verify mitigations. The strategies to reduce transient stimulation artifacts and improve performance during aDBS were confirmed in a chronic implant using updated configuration settings. Main results. We derived and validated a ‘checklist’ of configuration settings to improve system performance and areas for future device improvement. Key considerations for the configuration include (a) active instead of passive recharge, (b) sense-channel blanking in the amplifier, (c) high-pass filter settings, (d) tissue–electrode impedance mismatch management, (e) time-frequency trade-offs in the classifier, (f) algorithm blanking and transition rate limits. Without proper channel configuration, the aDBS algorithm was susceptible to limit-cycles of oscillating stimulation independent of physiological state. By applying the checklist, we could optimize each block’s performance characteristics within the overall system. With system-level optimization, a ‘fast’ aDBS prototype algorithm was demonstrated to be feasible without reentrant loops, and with noise performance suitable for subcortical brain circuits. Significance. We present a framework to study sources and propose mitigations of artifacts in devices that provide chronic aDBS. This work highlights the trade-offs in performance as novel sensing devices translate to the clinic. Finding the appropriate balance of constraints is imperative for successful translation of aDBS therapies. Clinical trial: Institutional Review Board and Investigational Device Exemption numbers: NCT02649166/IRB201501021 (University of Florida), NCT04043403/IRB52548 (Stanford University), NCT03582891/IRB1824454 (University of California San Francisco). IDE #180 097.

Motor fluctuations in Parkinson’s disease are characterized by unpredictability in the timing and duration of dopaminergic therapeutic benefits on symptoms, including bradykinesia and rigidity. These fluctuations significantly impair the quality of life of many Parkinson’s patients. However, current clinical evaluation tools are not designed for the continuous, naturalistic (real-world) symptom monitoring needed to optimize clinical therapy to treat fluctuations. Although commercially available wearable motor monitoring, used over multiple days, can augment neurological decision making, the feasibility of rapid and dynamic detection of motor fluctuations is unclear. So far, applied wearable monitoring algorithms are trained on group data. In this study, we investigated the influence of individual model training on short timescale classification of naturalistic bradykinesia fluctuations in Parkinson’s patients using a single-wrist accelerometer. As part of the Parkinson@Home study protocol, 20 Parkinson patients were recorded with bilateral wrist accelerometers for a one hour OFF medication session and a one hour ON medication session during unconstrained activities in their own homes. Kinematic metrics were extracted from the accelerometer data from the bodyside with the largest unilateral bradykinesia fluctuations across medication states. The kinematic accelerometer features were compared over the 1 h duration of recording, and medication-state classification analyses were performed on 1 min segments of data. Then, we analyzed the influence of individual versus group model training, data window length, and total number of training patients included in group model training, on classification. Statistically significant areas under the curves (AUCs) for medication induced bradykinesia fluctuation classification were seen in 85% of the Parkinson patients at the single minute timescale using the group models. Individually trained models performed at the same level as the group trained models (mean AUC both 0.70, standard deviation respectively 0.18 and 0.10) despite the small individual training dataset. AUCs of the group models improved as the length of the feature windows was increased to 300 s, and with additional training patient datasets. We were able to show that medication-induced fluctuations in bradykinesia can be classified using wrist-worn accelerometry at the time scale of a single minute. Rapid, naturalistic Parkinson motor monitoring has the clinical potential to evaluate dynamic symptomatic and therapeutic fluctuations and help tailor treatments on a fast timescale.

Invasive neural recording in humans shows promise for understanding the circuit basis of brain disorders. Most recordings have been done for short durations from externalized brain leads in hospital settings, or from first-generation implantable sensing devices that offer only intermittent brief streaming of time series data. Here we report the first human use of an implantable neural interface for wireless multichannel streaming of field potentials over long periods, with and without simultaneous therapeutic neurostimulation, untethered to receiving devices. Four Parkinson’s disease patients streamed bilateral 4-channel motor cortical and basal ganglia field potentials at home for over 500 hours, paired with wearable monitors that behaviorally categorize states of inadequate or excessive movement. Motor state during normal home activities was efficiently decoded using either supervised learning or unsupervised clustering algorithms. This platform supports adaptive deep brain stimulation, and may be widely applicable to brain disorders treatable by invasive neuromodulation.