Search

Excessive daytime sleepiness is a recognized non-motor symptom that adversely impacts the quality of life of people with Parkinson's disease (PD), yet effective treatment options remain limited. Deep brain stimulation (DBS) of the subthalamic nucleus (STN) is an effective treatment for PD motor signs. Reliable daytime sleep-wake classification using local field potentials (LFPs) recorded from DBS leads implanted in STN can inform the development of closed-loop DBS approaches for prompt detection and disruption of sleep-related neural oscillations. We performed STN DBS lead recordings in three nonhuman primates rendered parkinsonian by administrating neurotoxin 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP). Reference sleep-wake states were determined on a second-by-second basis by video monitoring of eyes (eyes-open, wake and eyes-closed, sleep). The spectral power in delta (1-4 Hz), theta (4-8 Hz), low-beta (8-20 Hz), high-beta (20-35 Hz), gamma (35-90 Hz), and high-frequency (200-400 Hz) bands were extracted from each wake and sleep epochs for training (70% data) and testing (30% data) a support vector machines classifier for each subject independently. The spectral features yielded reasonable daytime sleep-wake classification (sensitivity: 90.68 ± 1.28; specificity: 88.16 ± 1.08; accuracy: 89.42 ± 0.68; positive predictive value; 88.70 ± 0.89, n = 3). Our findings support the plausibility of monitoring daytime sleep-wake states using DBS lead recordings. These results could have future clinical implications in informing the development of closed-loop DBS approaches for automatic detection and disruption of sleep-related neural oscillations in people with PD to promote wakefulness.

Rigidity of upper and lower limbs in Parkinson's disease (PD) is typically assessed via a clinical rating scale that is subject to human perception biases. Methodologies to quantify changes in rigidity associated with the angular position (stiffness) or velocity (viscous damping) are needed to enhance our understanding of PD pathophysiology and objectively assess therapies. In this proof of concept study, we developed a robotic system and a model-based approach to estimate viscous damping and stiffness of the elbow. Our methodology enables the subject to freely rotate the elbow using an admittance controller while torque perturbations tailored to identify the arm dynamics are delivered. The viscosity and stiffness are calculated based on the experimental data using least-squares estimation. We validated our technique using computer simulations and experiments with a nonhuman animal model of PD in the presence and absence of deep brain stimulation therapy. Our data show that stiffness and viscosity measurements can better differentiate rigidity changes than scores previously used for research, including the work and impulse scores, and the modified unified Parkinson's disease rating scale. Our estimation method is suitable for quantifying the effect of therapies on viscous damping and stiffness and studying the pathophysiological mechanisms underlying rigidity in PD.

To elucidate the role of the basal ganglia during REM sleep movements in Parkinson’s disease (PD) we recorded pallidal neural activity from four PD patients. Unlike desynchronization commonly observed during wakeful movements, beta oscillations (13-35 Hz) synchronized during REM sleep movements; furthermore, high-frequency oscillations (150-350 Hz) synchronized during movement irrespective of sleep-wake states. Our results demonstrate differential engagement of the basal ganglia during REM sleep and awake movements.

Deep brain stimulation (DBS) surgery has been shown to dramatically improve the quality of life for patients with various motor dysfunctions, such as those afflicted with Parkinson's disease (PD), dystonia, and essential tremor (ET), by relieving motor symptoms associated with such pathologies. The success of DBS procedures is directly related to the proper placement of the electrodes, which requires the ability to accurately detect and identify relevant target structures within the subcortical basal ganglia region. In particular, accurate and reliable segmentation of the globus pallidus (GP) interna is of great interest for DBS surgery for PD and dystonia. In this study, we present a deep‐learning based neural network, which we term GP‐net, for the automatic segmentation of both the external and internal segments of the globus pallidus. High resolution 7 Tesla images from 101 subjects were used in this study; GP‐net is trained on a cohort of 58 subjects, containing patients with movement disorders as well as healthy control subjects. GP‐net performs 3D inference in a patient‐specific manner, alleviating the need for atlas‐based segmentation. GP‐net was extensively validated, both quantitatively and qualitatively over 43 test subjects including patients with movement disorders and healthy control and is shown to consistently produce improved segmentation results compared with state‐of‐the‐art atlas‐based segmentations. We also demonstrate a postoperative lead location assessment with respect to a segmented globus pallidus obtained by GP‐net., Deep brain stimulation (DBS) surgery has been shown to improve the quality of life for patients with various motor dysfunctions. The success of DBS is directly related to the proper placement of the electrodes, which requires accurate detection and identification of the relevant target structures. We present a deep‐learning based automatic, robust and accurate segmentation technique from 7 Tesla MRI acquisitions of the globus pallidus externa and interna for DBS surgery planning and postoperative electrode localization.

Background Assessing postural stability in Parkinson’s disease (PD) often relies on measuring the stepping response to an imposed postural perturbation. The standard clinical technique relies on a brisk backwards pull at the shoulders by the examiner and judgement by a trained rater. In research settings, various quantitative measures and perturbation directions have been tested, but it is unclear which metrics and perturbation direction differ most between people with PD and controls. Objectives (1) Use standardized forward vs. backward perturbations of a support surface to evaluate reactive stepping performance between PD and control participants. (2) Evaluate the utility of using principal components analysis to capture the dynamics of the reactive response and differences between groups. Methods Sixty-two individuals participated (40 mild-to-moderate PD, off medication). Standardized rapid translations of the support surface were applied, requiring at least one step, backward or forward, to restore balance. The number of steps taken and the projection of the first principal component (PC1) of the center of pressure (COP) time series were entered in linear repeated-measures mixed effect models. Results Forward falls required significantly fewer steps to recover than backward falls. PC1 captured more than half of the variance in the COP trajectory. Analysis of the PC1 projection revealed a significant interaction effect of group (PD vs. controls) by direction, such that there was a group difference in forward stepping, but not backward. Significance Forward reactive stepping in PD differed from controls more than backward-stepping. PC1 projections of the COP trajectory capture the dynamics of the postural response and differ between PD and controls.

Precise neuromodulation systems are needed to identify the role of neural oscillatory dynamics in brain function, and to advance the development of electrical stimulation therapies tailored to each patient’s signature of brain dysfunction. Low-frequency, local field potentials (LFPs) are of increasing interest for the development of these systems because they reflect the synaptic inputs to a recorded neuronal population and can be chronically recorded in humans. Here, we identified stimulation pulse sequences to optimally minimize or maximize the 2-norm of frequency-specific LFP oscillations using a generalized mathematical model of spontaneous and stimulation-evoked LFP activity, and a subject-specific model of neural dynamics in the pallidum of a Parkinson’s disease patient. We leveraged convex and mixed-integer optimization tools to identify the pulse patterns, and employed constraints on the pulse frequency and amplitude that are required to keep electrical stimulation within its safety envelope. Our analysis revealed that a combination of phase, amplitude, and frequency pulse modulation is needed to attain optimal suppression or amplification of the targeted oscillations. Phase modulation is sufficient to modulate oscillations with a constant amplitude envelope. To attain optimal modulation for oscillations with a time-varying envelope, a trade-off between frequency and amplitude pulse modulation is needed. The optimized pulse sequences identified here were invariant to changes in the dynamics of stimulation-evoked neural activity, including the damping and natural frequency or complexity (i.e., generalized vs. patient-specific model). Our results reveal the structure of pulse sequences that can be used in closed-loop brain stimulation devices to control neural activity in real-time.Significance StatementThis computational study reveals the structure of electrical pulse sequences that optimally modulate spontaneous, synaptic-related neural oscillations measured using local field potentials (LFPs). We considered constraints on the pulse amplitude and frequency, relevant to keep stimulation pulses within the safety envelope. Our mathematical analysis indicate that these optimized pulse sequences are invariant to changes in the studied neural dynamics, and a trade-off between amplitude and frequency pulse modulation is critical when the pulse amplitude is constrained. The pulse sequences identified here provide the rationale for the development of closed-loop neuromodulation systems that employ stimulation-evoked responses to control neural activity. These closed-loop systems will be key to characterize the role of neural activity in brain function and ultimately advance personalized neuromodulation devices.

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) and globus pallidus internus (GPi) is an effective treatment for parkinsonian motor signs. Though its therapeutic mechanisms remain unclear, it has been suggested that antidromic activation of the primary motor cortex (M1) plays a significant role in mediating its therapeutic effects. This study tested the hypothesis that antidromic activation of M1 is a prominent feature underlying the therapeutic effect of STN and GPi DBS. Single-unit activity in M1 was recorded using high-density microelectrode arrays in two parkinsonian nonhuman primates each implanted with DBS leads targeting the STN and GPi. Stimulation in each DBS target had similar therapeutic effects, however, antidromic activation of M1 was only observed during STN DBS. Although both animals undergoing STN DBS had similar beneficial effects, the proportion of antidromic-classified cells in each differed, 30 versus 6%. Over 4 h of continuous STN DBS, antidromic activation became less robust, whereas therapeutic benefits were maintained. Although antidromic activation waned over time, synchronization of spontaneous spiking in M1 was significantly reduced throughout the 4 h. Although we cannot discount the potential therapeutic role of antidromic M1 activation at least in the acute phase of STN DBS, the difference in observed antidromic activation between animals, and target sites, raise questions about its hypothesized role as the primary mechanism underlying the therapeutic effect of DBS. These results lend further support that reductions in synchronization at the level of M1 are an important factor in the therapeutic effects of DBS.SIGNIFICANCE STATEMENTRecently there has been great interest and debate regarding the potential role of motor cortical activation in the therapeutic mechanisms of deep brain stimulation (DBS) for Parkinson's disease. In this study we used chronically implanted high density microelectrode arrays in primary motor cortex (M1) to record neuronal population responses in parkinsonian nonhuman primates during subthalamic nucleus (STN) DBS and globus pallidus internus (GPi) DBS. Our results suggest a contribution of antidromic activation of M1 during STN DBS in disrupting synchronization in cortical neuronal populations; however, diminishing antidromic activity over time, and differences in observed antidromic activation between animals and target sites with antidromic activation not observed during GPi DBS, raise questions about its role as the primary mechanism underlying the therapeutic effect of DBS.

Parkinson’s disease (PD) is a progressive neurodegenerative movement disorder affecting over 10 million people worldwide. In the 1930s and 1940s there was little understanding regarding what caused PD or how to treat it. In a desperate attempt to improve patients’ lives different regions of the neuraxis were ablated. Morbidity and mortality were common, but some patients’ motor signs improved with lesions involving the basal ganglia or thalamus. With the discovery of l -dopa the advent of medical therapy began and surgical approaches became less frequent. It soon became apparent, however, that medical therapy was associated with side effects in the form of drug-induced dyskinesia and motor fluctuations and surgical therapies reemerged. Fortunately, during this time studies in monkeys had begun to lay the groundwork to understand the functional organization of the basal ganglia, and with the discovery of the neurotoxin MPTP a monkey model of PD had been developed. Using this model scientists were characterizing the physiological changes that occurred in the basal ganglia in PD and models of basal ganglia function and dysfunction were proposed. This work provided the rationale for the return of pallidotomy, and subsequently deep brain stimulation procedures. In this paper we describe the evolution of these monkey studies, how they provided a greater understanding of the pathophysiology underlying the development of PD and provided the rationale for surgical procedures, the search to understand mechanisms of DBS, and how these studies have been instrumental in understanding PD and advancing the development of surgical therapies for its treatment.

Parkinson's disease (PD) may optimally be treated with a disease-modifying therapy to slow progression. We compare data underlying surgical approaches proposed to impart disease modification in PD: (1) cell transplantation therapy with stem cell-derived dopaminergic neurons to replace damaged cells; (2) clinical trials of growth factors to promote survival of existing dopaminergic neurons; (3) subthalamic nucleus deep brain stimulation early in the course of PD; and (4) abdominal vagotomy to lower risk of potential disease spread from gut to brain. Though targeted to engage potential mechanisms of PD these surgical approaches remain experimental, indicating the difficulty in translating therapeutic concepts into clinical practice. The choice of outcome measures to assess disease modification separate from the symptomatic benefit will be critical to evaluate the effect of the disease-modifying intervention on long-term disease burden, including imaging studies and clinical rating scales, i.e., Unified Parkinson Disease Rating Scale. Therapeutic interventions will require long follow-up times (i.e., 5-10 years) to analyze disease modification compared to symptomatic treatments. The promise of invasive, surgical treatments to achieve disease modification through mechanistic approaches has been constrained by the reality of translating these concepts into effective clinical trials.

Coordinated reset deep brain stimulation (CR DBS) in the subthalamic nucleus (STN) has been demonstrated effective for the treatment of the motor signs associated with Parkinson's disease (PD). A critical CR parameter is an order in which stimulation is delivered across contacts. The relative effect of alternating vs. not alternating this order, i.e., shuffling vs. non-shuffling, however, has not been evaluated in vivo. The objective of this study is to compare the effect of shuffled vs. non-shuffled STN CR DBS on Parkinsonian motor signs. Two Parkinsonian non-human primates were implanted with a DBS lead in the STN. The effects of STN CR DBS with and without shuffling were compared with the traditional isochronal DBS (tDBS) using a within-subject design. For each stimulation setting, DBS was delivered for 2 or 4 h/day for 5 consecutive days. The severity of PD was assessed using a modified clinical rating scale immediately before, during, and 1 h after DBS, as well as on days following the discontinuation of the 5 days of daily stimulation, i.e., carryover effect. Shuffled STN CR DBS produced greater acute and carryover improvements on Parkinsonian motor signs compared with non-shuffled CR. Moreover, this difference was more pronounced when more effective stimulation intensity and burst frequency settings were used. tDBS showed limited carryover effects. Given the significant effect of shuffling on the effectiveness of CR DBS, it will be critical for future studies to further define the relative role of different CR parameters for the clinical implementation of this novel stimulation paradigm.

The purpose of this work is to present a new method that can be used to estimate and mitigate RF induced currents on Deep Brain Stimulation (DBS) leads. Here, we demonstrate the effect of RF induced current mitigation on both RF heating and image quality for a variety of brain MRI sequences at 3 T. We acquired pre-scan images around a DBS lead (in-situ and ex-vivo) using conventional Gradient Echo Sequence (GRE) accelerated by parallel imaging (i.e GRAPPA) and quantified the magnitude and phase of RF induced current using the relative location of the B1+ null with respect to the lead position. We estimated the RF induced current on a DBS lead implanted in a gel phantom as well as in a cadaver head study for a variety of RF excitation patterns. We also measured the increase in tip temperature using fiber-optic probes for both phantom and cadaver studies. Using the magnitude and phase information of the current induced separately by two transmit channels of the body coil, we calculated an implant friendly (IF) excitation. Using the IF excitation, we acquired T1, T2 weighted Turbo Spin Echo (TSE), T2 weighted SPACE-Dark Fluid, and Ultra Short Echo Time (UTE) sequences around the lead. Our induced current estimation demonstrated linear relationship between the magnitude of the induced current and the square root SAR at the tip of the lead as measured in phantom studies. The “IF excitation pattern” calculated after the pre-scan mitigated RF artifacts and increased the image quality around the lead. In addition, it reduced the tip temperature significantly in both phantom and cadaver studies compared to a conventional quadrature excitation while keeping equivalent overall image quality. We present a relatively fast method that can be used to calculate implant friendly excitation, reducing image artifacts as well as the temperature around the DBS electrodes. When combined with a variety of MR sequences, the proposed method can improve the image quality and patient safety in clinical imaging scenarios.

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) has shown clinical potential for relieving the motor symptoms of advanced Parkinson’s disease. While accurate localization of the STN is critical for consistent across-patients effective DBS, clear visualization of the STN under standard clinical MR protocols is still challenging. Therefore, intraoperative microelectrode recordings (MER) are incorporated to accurately localize the STN. However, MER require significant neurosurgical expertise and lengthen the surgery time. Recent advances in 7 T MR technology facilitate the ability to clearly visualize the STN. The vast majority of centers, however, still do not have 7 T MRI systems, and fewer have the ability to collect and analyze the data. This work introduces an automatic STN localization framework based on standard clinical MRIs without additional cost in the current DBS planning protocol. Our approach benefits from a large database of 7 T MRI and its clinical MRI pairs. We first model in the 7 T database, using efficient machine learning algorithms, the spatial and geometric dependency between the STN and its adjacent structures (predictors). Given a standard clinical MRI, our method automatically computes the predictors and uses the learned information to predict the patient-specific STN. We validate our proposed method on clinical T(2)W MRI of 80 subjects, comparing with experts-segmented STNs from the corresponding 7 T MRI pairs. The experimental results show that our framework provides more accurate and robust patient-specific STN localization than using state-of-the-art atlases. We also demonstrate the clinical feasibility of the proposed technique assessing the post-operative electrode active contact locations.

Background Abnormal oscillatory neural activity in the beta-frequency band (13-35 Hz) is thought to play a role in Parkinson's disease (PD); however, increasing evidence points to alterations in high-frequency ranges (>100 Hz) also having pathophysiological relevance. Objectives Studies have found that power in subthalamic nucleus (STN) high-frequency oscillations is increased with dopaminergic medication and during voluntary movements, implicating these brain rhythms in normal basal ganglia function. The objective of this study was to investigate whether similar signaling occurs in the internal globus pallidus (GPi), a nucleus increasingly used as a target for deep brain stimulation (DBS) for PD. Methods Spontaneous and movement-related GPi field potentials were recorded from DBS leads in 5 externalized PD patients on and off dopaminergic medication, as well as from 3 rhesus monkeys before and after the induction of parkinsonism with the neurotoxin 1-methyl-4-phenyl-1,2,3,6 tetrahydropyridine. Results In the parkinsonian condition, we identified a prominent oscillatory peak centered at 200-300 Hz that increased during movement. In patients the magnitude of high-frequency oscillation modulation was negatively correlated with bradykinesia. In monkeys, high-frequency oscillations were mostly absent in the naive condition but emerged after the neurotoxin 1-methyl-4-phenyl-1,2,3,6 tetrahydropyridine. In patients, spontaneous high-frequency oscillations were significantly attenuated on-medication. Conclusions Our findings provide evidence in support of the hypothesis that exaggerated, movement-modulated high-frequency oscillations in the GPi are pathophysiological features of PD. These findings suggest that the functional role(s) of high-frequency oscillations may differ between the STN and GPi and motivate additional investigations into their relationship to motor control in normal and diseased states.

Background: Rigidity of upper and lower limbs in Parkinson9s disease (PD) is typically assessed via a clinical rating scale that is subject to biases inherent to human perception. Methodologies and systems to systematically quantify changes in rigidity associated with angular position (stiffness) or velocity (viscous damping) are needed to enhance our understanding of PD pathophysiology and objectively assess therapies. New method: We developed a robotic system and a model-based approach to estimate viscous damping and stiffness of the elbow. Our methodology enables the subject to freely rotate the elbow, while torque perturbations tailored to identify the arm dynamics are delivered. The viscosity and stiffness are calculated based on the experimental data using least-squares optimization. Results: We validated our technique using computer simulations of the arm dynamics and experiments with a nonhuman primate model of PD in the presence and absence of deep brain stimulation (DBS) therapy. Computer simulations of the arm demonstrate that the proposed approach can accurately estimate viscous damping and stiffness. The experimental data show that both viscosity and stiffness significantly decreased when DBS was delivered. Comparison with existing methods: Computer simulations and experiments showed that stiffness and viscosity measurements obtained with the proposed methodology could better differentiate changes in rigidity than scores previously used for research, including the work score, impulse score, and modified Unified Parkinson9s Disease Rating Scale (mUPDRS). Conclusions: The proposed estimation method is suitable to quantify the effect of therapies on viscous damping and stiffness and study the pathophysiological mechanisms underlying rigidity in PD.

Subthalamic nucleus (STN) deep brain stimulation (DBS) is an established therapy for Parkinson’s disease motor symptoms. The ideal site for implantation within STN, however, remains controversial. While many argue that placement of a DBS lead within the sensorimotor territory of the STN yields better motor outcomes, others report similar effects with leads placed in the associative or motor territory of the STN, while still others assert that placing a DBS lead “anywhere within a 6-mm-diameter cylinder centered at the presumed middle of the STN (based on stereotactic atlas coordinates) produces similar clinical efficacy.” These discrepancies likely result from methodological differences including targeting preferences, imaging acquisition and the use of brain atlases that do not account for patient-specific anatomic variability. We present a first-in-kind within-patient demonstration of severe mood side effects and minimal motor improvement in a Parkinson’s disease patient following placement of a DBS lead in the limbic/associative territory of the STN who experienced marked improvement in motor benefit and resolution of mood side effects following repositioning the lead within the STN sensorimotor territory. 7 Tesla (7 T) magnetic resonance imaging (MRI) data were used to generate a patient-specific anatomical model of the STN with parcellation into distinct functional territories and computational modeling to assess the relative degree of activation of motor, associative and limbic territories.

Background Deep brain stimulation (DBS) of the pedunculopontine nucleus (PPN) has been reported to improve gait disturbances in Parkinson's disease (PD); however, there are controversies on the radiological and electrophysiological techniques for intraoperative and postoperative confirmation of the target and determination of optimal stimulation parameters. Objectives We investigated the correlation between the location of the estimated PPN (ePPN) and neuronal activity collected during intraoperative electrophysiological mapping to evaluate the role of microelectrode recording (MER) in identifying the effective stimulation site in two PD patients. Materials and methods Bilateral PPN DBS was performed in two patients who had suffered from levodopa refractory gait disturbance. They had been implanted previously with DBS in the internal globus pallidus and the subthalamic nucleus, respectively. The PPN was determined on MRI and identified by intraoperative MER. Neuronal activity recorded was analyzed for mean discharge rate, bursting, and oscillatory activity. The effects were assessed by clinical ratings for motor signs before and after surgery. Results The PPN location was detected by MER. Groups of neurons characterized by tonic discharges were found 9-10 mm below the thalamus. The mean discharge rate in the ePPN was 19.1 ± 15.1 Hz, and 33% of the neurons of the ePPN responded with increased discharge rate during passive manipulation of the limbs and orofacial structures. PPN DBS with bipolar stimulation at a frequency range 10-30 Hz improved gait disturbances in both patients. Although PPN DBS provided therapeutic effects post-surgery in both cases, the effects waned after a year in case 1 and three years in case 2. Conclusions Estimation of stimulation site within the PPN is possible by combining physiological guidance using MER and MRI findings. The PPN is a potential target for gait disturbances, although the efficacy of PPN DBS may depend on the location of the electrode and the stimulation parameters.

Objective. The electrode–tissue interface surrounding a deep brain stimulation (DBS) lead is known to be highly dynamic following implantation, which may have implications on the interpretation of intraoperatively recorded local field potentials (LFPs). We characterized beta-band LFP dynamics following implantation of a directional DBS lead in the sensorimotor subthalamic nucleus (STN), which is a primary target for treating Parkinson’s disease. Approach. Directional STN-DBS leads were implanted in four healthy, non-human primates. LFPs were recorded over two weeks and again 1–4 months after implantation. Impedance was measured for two weeks post-implant without stimulation to compare the reactive tissue response to changes in LFP oscillations. Beta-band (12–30 Hz) peak power was calculated from the LFP power spectra using both common average referencing (CAR) and intra-row bipolar referencing (IRBR). Results. Resting-state LFPs in two of four subjects revealed a steady increase of beta power over the initial two weeks post-implant whereas the other two subjects showed variable changes over time. Beta power variance across days was significantly larger in the first two weeks compared to 1–4 months post-implant in all three long-term subjects. Further, spatial maps of beta power several hours after implantation did not correlate with those measured two weeks or 1–4 months post-implant. CAR and IRBR beta power correlated across short- and long-term time points. However, depending on the time period, subjects showed a significant bias towards larger beta power using one referencing scheme over the other. Lastly, electrode–tissue impedance increased over the two weeks post-implant but showed no significant correlation to beta power. Significance. These results suggest that beta power in the STN may undergo significant changes following DBS lead implantation. DBS lead diameter and electrode recording configurations can affect the post-implant interpretation of oscillatory features. Such insights will be important for extrapolating results from intraoperative and externalized LFP recordings.

Deep brain stimulation (DBS) surgery has been shown to dramatically improve the quality of life for patients with various motor dysfunctions, such as those afflicted with Parkinson’s disease (PD), dystonia, and essential tremor (ET), by relieving motor symptoms associated with such pathologies. The success of DBS procedures is directly related to the proper placement of the electrodes, which requires the ability to accurately detect and identify relevant target structures within the subcortical basal ganglia region. In particular, accurate and reliable segmentation of the globus pallidus (GP) interna is of great interest for DBS surgery for PD and dystonia. In this work, we present a deep-learning based neural network, which we term GP-net, for the automatic segmentation of both the external and internal segments of the globus pallidus. High resolution 7 Tesla images from 101 subjects were used in this study; GP-net is trained on a cohort of 58 subjects, containing patients with movement disorders as well as healthy control subjects. GP-net performs 3D inference in a patient-specific manner, alleviating the need for atlas-based segmentation. GP-net was extensively validated, both quantitatively and qualitatively over 43 test subjects including patients with movement disorders and healthy control and is shown to consistently produce improved segmentation results compared with state-of-the-art atlas-based segmentations. We also demonstrate a post-operative lead location assessment with respect to a segmented globus pallidus obtained by GP-net.