Comparison of state-of-the-art deep learning architectures for detection of freezing of gait in Parkinson's disease

Introduction

Freezing of gait (FOG) is one of the most debilitating motor symptoms experienced by patients with Parkinson's disease (PD). FOG detection is possible using acceleration data from wearable sensors, and a convolutional neural network (CNN) is often used to determine the presence of FOG epochs. We compared the performance of a standard CNN for the detection of FOG with two more complex networks, which are well suited for time series data, the MiniRocket and the InceptionTime.

Methods

We combined acceleration data of people with PD across four studies. The final data set was split into a training (80%) and hold-out test (20%) set. A fifth study was included as an unseen test set. The data were windowed (2 s) and five-fold cross-validation was applied. The CNN, MiniRocket, and InceptionTime models were evaluated using a receiver operating characteristic (ROC) curve and its area under the curve (AUC). Multiple sensor configurations were evaluated for the best model. The geometric mean was subsequently calculated to select the optimal threshold. The selected model and threshold were evaluated on the hold-out and unseen test set.

Results

A total of 70 participants (23.7 h, 9% FOG) were included in this study for training and testing, and in addition, 10 participants provided an unseen test set (2.4 h, 11% FOG). The CNN performed best (AUC = 0.86) in comparison to the InceptionTime (AUC = 0.82) and MiniRocket (AUC = 0.76) models. For the CNN, we found a similar performance for a seven-sensor configuration (lumbar, upper and lower legs and feet; AUC = 0.86), six-sensor configuration (upper and lower legs and feet; AUC = 0.87), and two-sensor configuration (lower legs; AUC = 0.86). The optimal threshold of 0.45 resulted in a sensitivity of 77% and a specificity of 58% for the hold-out set (AUC = 0.72), and a sensitivity of 85% and a specificity of 68% for the unseen test set (AUC = 0.90).

Conclusion

We confirmed that deep learning can be used to detect FOG in a large, heterogeneous dataset. The CNN model outperformed more complex networks. This model could be employed in future personalized interventions, with the ultimate goal of using automated FOG detection to trigger real-time cues to alleviate FOG in daily life.

Predicting Motor Outcome and Quality of Life After Subthalamic Deep Brain Stimulation for Parkinson's Disease: The Role of Standard Screening Measures and Wearable-Data

BACKGROUND

Standardized screening for subthalamic deep brain stimulation (STN DBS) in Parkinson's disease (PD) patients is crucial to determine eligibility, but its utility to predict postoperative outcomes in eligible patients is inconclusive. It is unknown whether wearable data can contribute to this aim.

OBJECTIVE

To evaluate the utility of universal components incorporated in the DBS screening, complemented by a wearable sensor, to predict motor outcomes and Quality of life (QoL) one year after STN DBS surgery.

METHODS

Consecutive patients were included in the OPTIMIST cohort study from two DBS centers. Standardized assessments included a preoperative Levodopa Challenge Test (LCT), and questionnaires on QoL and non-motor symptoms including cognition, psychiatric symptoms, impulsiveness, autonomic symptoms, and sleeping problems. Moreover, an ambulatory wearable sensor (Parkinson Kinetigraph (PKG)) was used. Postoperative assessments were similar and also included a Stimulation Challenge Test to determine DBS effects on motor function.

RESULTS

Eighty-three patients were included (median (interquartile range) age 63 (56-68) years, 36% female). Med-OFF (Stim-OFF) motor severity deteriorated indicating disease progression, but patients significantly improved in terms of Med-ON (Stim-ON) motor function, motor fluctuations, QoL, and most non-motor domains. Motor outcomes were not predicted by preoperative tests, including covariates of either LCT or PKG. Postoperative QoL was predicted by better preoperative QoL, lower age, and more preoperative impulsiveness scores in multivariate models.

CONCLUSION

Data from the DBS screening including wearable data do not predict postoperative motor outcome at one year. Post-DBS QoL appears primarily driven by non-motor symptoms, rather than by motor improvement.

Involving Patients in Weighting Benefits and Harms of Treatment in Parkinson's Disease

Introduction

Little is known about how patients weigh benefits and harms of available treatments for Parkinson’s Disease (oral medication, deep brain stimulation, infusion therapy). In this study we have (1) elicited patient preferences for benefits, side effects and process characteristics of treatments and (2) measured patients’ preferred and perceived involvement in decision-making about treatment.

Methods

Preferences were elicited using a best-worst scaling case 2 experiment. Attributes were selected based on 18 patient-interviews: treatment modality, tremor, slowness of movement, posture and balance problems, drowsiness, dizziness, and dyskinesia. Subsequently, a questionnaire was distributed in which patients were asked to indicate the most and least desirable attribute in nine possible treatment scenarios. Conditional logistic analysis and latent class analysis were used to estimate preference weights and identify subgroups. Patients also indicated their preferred and perceived degree of involvement in treatment decision-making (ranging from active to collaborative to passive).

Results

Two preference patterns were found in the patient sample (N = 192). One class of patients focused largely on optimising the process of care, while the other class focused more on controlling motor-symptoms. Patients who had experienced advanced treatments, had a shorter disease duration, or were still employed were more likely to belong to the latter class. For both classes, the benefits of treatment were more influential than the described side effects. Furthermore, many patients (45%) preferred to take the lead in treatment decisions, however 10.8% perceived a more passive or collaborative role instead.

Discussion

Patients weighted the benefits and side effects of treatment differently, indicating there is no “one-size-fits-all” approach to choosing treatments. Moreover, many patients preferred an active role in decision-making about treatment. Both results stress the need for physicians to know what is important to patients and to share treatment decisions to ensure that patients receive the treatment that aligns with their preferences.

Asymmetries in reactive and anticipatory balance control are of similar magnitude in Parkinson's disease patients

Many Parkinson's disease (PD) patients show asymmetries in balance control during quiet stance and in response to perturbations (i.e., reactive balance control) in the sagittal plane. In addition, PD patients show a reduced ability to anticipate to self-induced disturbances, but it is not clear whether these anticipatory responses can be asymmetric too. Furthermore, it is not known how reactive balance control and anticipatory balance control are related in PD patients. Therefore, we investigated whether reactive and anticipatory balance control are asymmetric to the same extent in PD patients. 14 PD patients and 10 controls participated. Reactive balance control (RBC) was investigated by applying external platform and force perturbations and relating the response of the left and right ankle torque to the body sway angle at the excited frequencies. Anticipatory postural adjustments (APAs) were investigated by determining the increase in the left and right ankle torque just before the subjects released a force exerted with the hands against a force sensor. The symmetry ratio between the contribution of the left and right ankle was used to express the asymmetry in reactive and anticipatory balance control; the correlation between the two ratio's was investigated with Spearman's rank correlation coefficients. PD patients were more asymmetric in anticipatory (p=0.026) and reactive balance control (p=0.004) compared to controls and the symmetry ratios were significantly related (ρ=0.74; p=0.003) in PD patients. These findings suggest that asymmetric reactive balance control during bipedal stance may share a common pathophysiology with asymmetries in the anticipation of voluntary perturbations during, for instance, gait initiation.

Balance asymmetry in Parkinson's disease and its contribution to freezing of gait

Balance control (the ability to maintain an upright posture) is asymmetrically controlled in a proportion of patients with Parkinson's disease. Gait asymmetries have been linked to the pathophysiology of freezing of gait. We speculate that asymmetries in balance could contribute to freezing by a) hampering the unloading of the stepping leg and/or b) leading to a preferred stance leg during gait, which then results in asymmetric gait. To investigate this, we examined the relationship between balance control and weight-bearing asymmetries and freezing. We included 20 human patients with Parkinson (tested OFF medication; nine freezers) and nine healthy controls. Balance was perturbed in the sagittal plane, using continuous multi-sine perturbations, applied by a motion platform and by a force at the sacrum. Applying closed-loop system identification techniques, relating the body sway angle to the joint torques of each leg separately, determined the relative contribution of each ankle and hip joint to the total amount of joint torque. We also calculated weight-bearing asymmetries. We determined the 99-percent confidence interval of weight-bearing and balance-control asymmetry using the responses of the healthy controls. Freezers did not have larger asymmetries in weight bearing (p = 0.85) nor more asymmetrical balance control compared to non-freezers (p = 0.25). The healthy linear one-to-one relationship between weight bearing and balance control was significantly different for freezers and non-freezers (p = 0.01). Specifically, non-freezers had a significant relationship between weight bearing and balance control (p = 0.02), whereas this relation was not significant for freezers (p = 0.15). Balance control is asymmetrical in most patients (about 75 percent) with Parkinson's disease, but this asymmetry is not related to freezing. The relationship between weight bearing and balance control seems to be less pronounced in freezers, compared to healthy controls and non-freezers. However, this relationship should be investigated further in larger groups of patients.

Subthalamic nucleus versus globus pallidus bilateral deep brain stimulation for advanced Parkinson's disease (NSTAPS study): a randomised controlled trial

BACKGROUND

Patients with advanced Parkinson's disease often have rapid swings between mobility and immobility, and many respond unsatisfactorily to adjustments in pharmacological treatment. We assessed whether globus pallidus pars interna (GPi) deep brain stimulation (DBS) gives greater functional improvement than does subthalamic nucleus (STN) DBS.

METHODS

We recruited patients from five centres in the Netherlands who were aged 18 years or older, had idiopathic Parkinson's disease, and had, despite optimum pharmacological treatment, at least one of the following symptoms: severe response fluctuations, dyskinesias, painful dystonias, or bradykinesia. By use of a computer-generated randomisation sequence, we randomly assigned patients to receive either GPi DBS or STN DBS (1:1), applying a minimisation procedure according to drug use (levodopa equivalent dose <1000 mg vs ≥1000 mg) and treatment centre. Patients and study assessors (but not those who assessed adverse events) were masked to treatment allocation. We had two primary outcomes: functional health as measured by the weighted Academic Medical Center Linear Disability Scale (ALDS; weighted by time spent in the off phase and on phase) and a composite score for cognitive, mood, and behavioural effects up to 1 year after surgery. Secondary outcomes were symptom scales, activities of daily living scales, a quality-of-life questionnaire, the occurrence of adverse events, and drug use. We used the intention-to-treat principle for all analyses. This trial is registered with www.controlled-trials.com, number ISRCTN85542074.

FINDINGS

Between Feb 1, 2007, and March 29, 2011, we enrolled 128 patients, assigning 65 to GPi DBS and 63 to STN DBS. We found no statistically significant difference in either of our primary outcomes: mean change in weighted ALDS (3·0 [SD 14·5] in the GPi group vs 7·7 [23·2] in the STN group; p=0·28) and the number of patients with cognitive, mood, and behavioural side-effects (36 [58%] of 62 patients in the GPi group vs 35 [56%] of 63 patients in the STN group; p=0·94). Secondary outcomes showed larger improvements in off-drug phase in the STN group compared with the GPi group in the mean change in unified Parkinson's disease rating scale motor examination scores (20·3 [16·3] vs 11·4 [16·1]; p=0·03), the mean change in ALDS scores (20·3 [27·1] vs 11·8 [18·9]; p=0·04), and medication (mean levodopa equivalent drug reduction: 546 [SD 561] vs 208 [521]; p=0·01). We recorded no difference in the occurrence of adverse events between the two groups. Other secondary endpoints showed no difference between the groups.

INTERPRETATION

Although there was no difference in our primary outcomes, our findings suggest that STN could be the preferred target for DBS in patients with advanced Parkinson's disease.

FUNDING

Stichting Internationaal Parkinson Fonds, Prinses Beatrix Fonds, and Parkinson Vereniging.

Toward Automated, At-Home Assessment of Mobility Among Patients With Parkinson Disease, Using a Body-Worn Accelerometer

Objective. To develop an automated and objective method to assess mobility in Parkinson disease (PD) patients in daily-life settings and to investigate whether accelerometer-derived measures discriminate between PD and healthy controls as they walk and simulate activities of daily living (ADL). Methods. Healthy older adults (17) and patients with PD (22) wore a triaxial accelerometer on their lower back during short walks (validation study) and during a walk around the medical center to simulate daily activities (ADL simulation). The variability (consistency and rhythmicity) of stepping was assessed. The patients completed the walks before and after taking their anti-Parkinsonian medications. Frequency-based acceleration measures included dominant frequency, amplitude (strength of signal frequency), width (frequency dispersion), and slope (a combination reflecting amplitude and width) of the main frequency of the power spectral density in the 0.5- to 3.0-Hz band. A subset of the Unified Parkinson-Disease Rating Scale provided a clinical measure of gait impairment (UPDRS-Gait5). A PD patient and control wore the sensors for 3 days at home. Results. The width was larger, and the amplitude and slope were smaller in the PD patients compared to the controls in the validation study and ADL simulation ( P < .02). The width decreased, and the amplitude and slope increased when patients took anti-Parkinsonian medications ( P < .007). Significant correlations were observed between acceleration-derived measures and UPDRS-Gait5. The data obtained at home was similar to the clinic data. Conclusions. Frequency-derived measures are valid and sensitive estimates of stride-to-stride variability that can be used to assess the quality and consistency of walking in patients with PD in real-life settings.

Clinical and pharmacogenetic determinants for the discontinuation of non-ergoline dopamine agonists in Parkinson's disease

Objective

To identify determinants for the discontinuation of non-ergoline dopamine agonist (DA) treatment in patients with Parkinson’s disease (PD) and to identify genetic determinants in genes encoding dopamine receptor (DR)D2 and DRD3 in a exploratory analysis.

Methods

Patients included were first-time users of the non-ergoline DA ropinirole or pramipexole who had been diagnosed with PD before 2005. Treatment discontinuation was defined as a gap of 180 days or more between two refills of the DA. Non-genetic determinants for discontinuation were studied in the overall population, and genetic determinants [DRD2 141C Ins/Del, DRD2 (CA)_n STR, DRD2 TaqIA, DRD3 MscI single nucleotide polymorphism (SNP) and DRD3 MspI SNP] were studied in a subgroup. Cox proportional hazard analysis was used to estimate the hazard ratios (HR) for the discontinuation of non-ergoline DA treatment.

Results

The study population comprised 90 patients. Apomorphine use was associated with non-ergoline DA discontinuation, although the apomorphine group consisted only of three patients [HR 6.26; 95% confidence interval (CI) 1.85–21.2]. Daily levodopa dosages between 500 and 1000 mg were positively associated with discontinuation (HR 2.31; 95% CI 1.08–4.93). Included in the exploratory pharmacogenetic analysis were 38 patients. The absence of a 15× DRD2 CA repeat allele was significantly related with a decreased discontinuation of non-ergoline treatment (HR 0.23; 95% CI 0.07–0.81). The DRD3 MspI polymorphism showed a non-significant allele dose effect, suggestive of a causal relationship.

Conclusion

This study identified apomorphine use and levodopa dosages between 500 and 1000 mg as non-genetic and the 15× DRD2 CA repeat allele as genetic determinants for the discontinuation of non-ergoline DA treatment in patients with PD. More research is needed to replicate these findings.

Ambulatory monitoring of activities and motor symptoms in Parkinson's disease

Ambulatory monitoring of motor symptoms in Parkinsons disease (PD) can improve our therapeutic strategies, especially in patients with motor fluctuations. Previously published monitors usually assess only one or a few basic aspects of the cardinal motor symptoms in a laboratory setting. We developed a novel ambulatory monitoring system that provides a complete motor assessment by simultaneously analyzing current motor activity of the patient (e.g. sitting, walking) and the severity of many aspects related to tremor, bradykinesia, and hypokinesia. The monitor consists of a set of four inertial sensors. Validity of our monitor was established in seven healthy controls and six PD patients treated with deep brain stimulation (DBS) of the subthalamic nucleus. Patients were tested at three different levels of DBS treatment. Subjects were monitored while performing different tasks, including motor tests of the Unified Parkinsons Disease Rating Scale (UPDRS). Output of the monitor was compared to simultaneously recorded videos. The monitor proved very accurate in discriminating between several motor activities. Monitor output correlated well with blinded UPDRS ratings during different DBS levels. The combined analysis of motor activity and symptom severity by our PD monitor brings true ambulatory monitoring of a wide variety of motor symptoms one step closer..

Intracranial hypertension in 2 children with marfan syndrome

Two unrelated children with Marfan syndrome presented with recurrent intracranial hypertension. Both children complained of headache, nausea, and vomiting and one of them had papilledema. Both had increased cerebrospinal fluid pressure, and their complaints disappeared after lumbar puncture. Although severe headache has been reported in Marfan syndrome due to intracranial hypotension, this is to our knowledge the first report of intracranial hypertension in Marfan patients.

Discontinuation of ropinirole and pramipexole in patients with Parkinson's disease: clinical practice versus clinical trials

OBJECTIVE

To compare characteristics and incidence of discontinuation of Parkinson's disease (PD) patients starting ropinirole or pramipexole in clinical practice with data from randomised controlled clinical trials (RCTs).

METHODS

Included in the retrospective clinical-practice cohort were first-time users of ropinirole or pramipexole diagnosed with PD before 2005. Baseline characteristics and incidence of discontinuation were compared between the clinical-practice cohort and RCTs. Treatment discontinuation was defined as more than 180 days between two refills of ropinirole or pramipexole. The incidence of discontinuation in RCTs was based on the reported rate of discontinuation for any cause.

RESULTS

Included were 45 patients who started with ropinirole and 59 patients who started with pramipexole. Treatment was discontinued within 3 years in 51% (ropinirole) and 60% (pramipexole) of the patients. Ten RCTs with ropinirole and 12 with pramipexole were identified. Baseline characteristics did not differ between the clinical-practice cohort and RCTs. RCTs reported discontinuation rates comparable with those at the same timepoint in the clinical practice until 1 year of follow-up.

CONCLUSION

This study shows that the overall incidence of discontinuation of ropinirole and pramipexole between the patients in our clinical-practice cohort and patients in the RCTs was comparable for the short term. However for the long term, discontinuation in practice is possibly higher.

Pharmacogenetics of antiparkinsonian drug treatment: a systematic review

Pharmacotherapy is the mainstay in the treatment of Parkinson's disease and the armamentarium of drugs available for the therapy of this disease is still expanding. Anti-Parkinson's disease drugs are effective in reducing the physical symptoms, such as hypokinesia, bradykinesia, rigidity and tremor. However, there is a large interindividual variability in response to anti-Parkinson's disease drugs with respect to both drug efficacy and toxicity. It is thought that genetic variability in genes encoding drug-metabolizing enzymes, drug receptors and proteins involved in pathway signaling is an important factor in determining interindividual variability in drug response. Pharmacogenetics aims at identifying genetic markers associated with drug response. Ideally, knowledge of these genetic markers will enable us to predict an individual's drug response in terms of both efficacy and toxicity. The role of pharmacogenetics in the treatment of Parkinson's disease is relatively unexplored. Therefore, we aim to present a systematic review of the published pharmacogenetic studies in Parkinson's disease and to describe polymorphic genes of interest for future research.

Longitudinal evaluation of "presymptomatic" carriers of Huntington's disease

The authors evaluated motor, behavioral, and cognitive functioning over a 3-year period in 33 presymptomatic carriers for Huntington's disease and compared them with 73 noncarriers. Investigators blind to the participant's gene status utilized the Unified Huntington's Disease Rating Scale (UHDRS) and performed an extensive neuropsychological assessment (global cognitive, memory, language, psychomotor). Successive evaluations of motor and behavioral patterns showed inconsistencies. The rate of cognitive changes in carriers was similar to that in noncarriers. Commonly used tools are inadequate for detecting markers in preclinical Huntington's disease, limiting the design of therapeutic trials. Research should focus on tracking suitable endpoints combining clinical markers and biomarkers that change linearly with disease progression.

Objective assessment of motor slowness in Huntington's disease: clinical correlates and 2-year follow-up

Functional disability of patients with Huntington's disease (HD) is determined by impairment of voluntary motor function rather than the presence of chorea. However, only few attempts have been made to quantify this motor impairment. By using a simple reaction time paradigm, we measured the time needed for movement initiation (akinesia) and execution (bradykinesia) in 76 HD patients and 127 controls. Akinesia and bradykinesia were already evident in early stages and increased linearly with increasing disease stage. Quantified motor slowness correlated with clinical impairment of voluntary movements but also with cognitive impairment and medication use. In patients without severe cognitive impairment, quantified motor slowness reflected clinical motor impairment more purely. During 1.9 years follow-up (range, 0.8-3.8 years), quantified akinesia and bradykinesia progressed concomitantly with progression of clinical impairment of voluntary movements, cognition, and functional capacity. However, rate of change in motor slowness did not discriminate between patients whose disease stage remained stable and those whose disease stage progressed. We conclude that the reaction time paradigm may be used to quantify akinesia and bradykinesia in HD, at least in patients without severe cognitive impairment. Although reaction and movement times increased in time, these measures failed to detect functionally important changes during our follow-up period.

Cognitive and motor functioning in gene carriers for Huntington's disease: a baseline study

The aim of this study was to investigate the first changes in cognitive and motor functioning in Huntington's disease. Forty-six gene carriers, not clinically diagnosed for HD, were compared with 88 non-gene carriers. Gene carriers performed significantly worse on the Benton Visual Retention Test. This result was due to a minority of participants who had already developed cognitive impairment. Marginal differences appeared on the motor times of single reaction time measures after correction for motor signs. The findings are discussed in the context of inconsistencies in previous studies and underscore the need for longitudinal research.