Quantitative evaluation of the effects of subthalamic stimulation on gait in Parkinson's disease patients using principal component analysis

Classification of gait patterns between patients with Parkinson's disease and healthy controls using phase space reconstruction (PSR), empirical mode decomposition (EMD) and neural networks

Parkinson's disease (PD) is a common neurodegenerative disorder that affects human's quality of life, especially leading to locomotor deficits such as postural instability and gait disturbances. Gait signal is one of the best features to characterize and detect movement disorders caused by a malfunction in parts of the brain and nervous system of the patients with PD. Various classification approaches using spatiotemporal gait variables have been presented earlier to classify Parkinson's gait. In this study we propose a novel method for gait pattern classification between patients with PD and healthy controls, based upon phase space reconstruction (PSR), empirical mode decomposition (EMD) and neural networks. First, vertical ground reaction forces (GRFs) at specific positions of human feet are captured and then phase space is reconstructed. The properties associated with the gait system dynamics are preserved in the reconstructed phase space. Three-dimensional (3D) PSR together with Euclidean distance (ED) has been used. These measured parameters demonstrate significant difference in gait dynamics between the two groups and have been utilized to form a reference variable set. Second, reference variables are decomposed into Intrinsic Mode Functions (IMFs) using EMD, and the third IMFs are extracted and served as gait features. Third, neural networks are then used as the classifier to distinguish between patients with PD and healthy controls based on the difference of gait dynamics preserved in the gait features between the two groups. Finally, experiments are carried out on 93 PD patients and 73 healthy subjects to assess the effectiveness of the proposed method. By using 2-fold, 10-fold and leave-one-out cross-validation styles, the correct classification rates are reported to be 91.46%, 96.99% and 98.80%, respectively. Compared with other state-of-the-art methods, the results demonstrate superior performance and the proposed method can serve as a potential candidate for the automatic and non-invasive classification between patients with PD and healthy subjects.

A feature extraction method for adaptive DBS using an improved EMD

OBJECTIVE

Local field potential (LFP) of a patient with Parkinson's disease often shows abnormal oscillation phenomenon. Extracting and studying this phenomenon and designing adaptive deep brain stimulation (DBS) control library have great significance in the treatment of disease.

MATERIALS AND METHODS

This paper has designed a feature extraction method based on modified empirical mode decomposition (EMD) which extracts the abnormal oscillation signal in the time domain to increase the overall performance. The intrinsic mode function (IMF) component which contains abnormal oscillation is extracted by using EMD before an intrinsic characteristic of the oscillation signal is obtained. Abnormal oscillation signal is acquired using signal normalization, peak counting, and envelope method with a threshold which in turn keeps the integrity and accuracy as well as the efficiency.

RESULTS

Comparative study of eight patients (six patients with DBS closed and drugs stopped; two patients with stimulation) has verified the feasibility of using modified EMD in extracting abnormal oscillation signal. The results showed that patients who take DBS suffer less abnormal oscillation than those who take no treatment. These results match the energy rise in the band of 3-30 Hz on local field potential spectrum of the patient with Parkinson's disease.

CONCLUSIONS

Unlike previous oscillation extraction algorithm, improved EMD feature extraction method directly isolates abnormal oscillation signal from LFP. Significant improvement has been made in feature extraction algorithm in adaptability, real-time performance, and accuracy.

Shock attenuation characteristics of three different military boots during gait

Musculoskeletal injuries are related to the cushioning properties of boots in military populations. This study aimed to compare ground reaction force (GRF) and subjective perceived comfort from two different military boots supplied by the Brazilian Army with a commercial boot. Twenty army recruits volunteered for a GRF assessment during walking on a 10-m walkway and a perceived comfort test after 20min walking on a treadmill. Both experiments were conducted with three different military boots: CC10 (styrene-butadiene rubber - SBR - midsole 30mm thickness, 65 Shore A; 631.8g weight; supplied by the Brazilian Army); CC13 (SBR midsole 20.6mm thickness, 66 Shore A; 530.3g weight; supplied by the Brazilian Army) and CAT (polyurethane - PU - midsole 31.7mm thickness, 55 Shore A; 423g weight; commercially available). GRF was analyzed in the time (principal component analysis - PCA) and frequency (Blackman-Tukey) domains. No difference was found for the first and second peak forces or loading rate; however, significant influence from the military boots' design on GRF was found by PCA and frequency analysis. Loading factor presented higher values at early stance with lower force for CC10 compared to CC13 at these epochs. CC13 also presented higher power spectral density compared to CC10 at higher frequency bands. However, CAT was significantly more comfortable than CC10. These results suggest that the thicker SBR midsole boot was more effective in reducing impact, while the lightest boot with softer midsole hardness made with PU was the most comfortable.

Optimisation of a machine learning algorithm in human locomotion using principal component and discriminant function analyses

Assessment methods in human locomotion often involve the description of normalised graphical profiles and/or the extraction of discrete variables. Whilst useful, these approaches may not represent the full complexity of gait data. Multivariate statistical methods, such as Principal Component Analysis (PCA) and Discriminant Function Analysis (DFA), have been adopted since they have the potential to overcome these data handling issues. The aim of the current study was to develop and optimise a specific machine learning algorithm for processing human locomotion data. Twenty participants ran at a self-selected speed across a 15m runway in barefoot and shod conditions. Ground reaction forces (BW) and kinematics were measured at 1000 Hz and 100 Hz, respectively from which joint angles (°), joint moments (N.m.kg-1) and joint powers (W.kg-1) for the hip, knee and ankle joints were calculated in all three anatomical planes. Using PCA and DFA, power spectra of the kinematic and kinetic variables were used as a training database for the development of a machine learning algorithm. All possible combinations of 10 out of 20 participants were explored to find the iteration of individuals that would optimise the machine learning algorithm. The results showed that the algorithm was able to successfully predict whether a participant ran shod or barefoot in 93.5% of cases. To the authors' knowledge, this is the first study to optimise the development of a machine learning algorithm.

Immediate effects of EVA midsole resilience and upper shoe structure on running biomechanics: a machine learning approach

Background

Resilience of midsole material and the upper structure of the shoe are conceptual characteristics that can interfere in running biomechanics patterns. Artificial intelligence techniques can capture features from the entire waveform, adding new perspective for biomechanical analysis. This study tested the influence of shoe midsole resilience and upper structure on running kinematics and kinetics of non-professional runners by using feature selection, information gain, and artificial neural network analysis.

Methods

Twenty-seven experienced male runners (63 ± 44 km/week run) ran in four-shoe design that combined two resilience-cushioning materials (low and high) and two uppers (minimalist and structured). Kinematic data was acquired by six infrared cameras at 300 Hz, and ground reaction forces were acquired by two force plates at 1,200 Hz. We conducted a Machine Learning analysis to identify features from the complete kinematic and kinetic time series and from 42 discrete variables that had better discriminate the four shoes studied. For that analysis, we built an input data matrix of dimensions 1,080 (10 trials × 4 shoes × 27 subjects) × 1,254 (3 joints × 3 planes of movement × 101 data points + 3 vectors forces × 101 data points + 42 discrete calculated kinetic and kinematic features).

Results

The applied feature selection by information gain and artificial neural networks successfully differentiated the two resilience materials using 200(16%) biomechanical variables with an accuracy of 84.8% by detecting alterations of running biomechanics, and the two upper structures with an accuracy of 93.9%.

Discussion

The discrimination of midsole resilience resulted in lower accuracy levels than did the discrimination of the shoe uppers. In both cases, the ground reaction forces were among the 25 most relevant features. The resilience of the cushioning material caused significant effects on initial heel impact, while the effects of different uppers were distributed along the stance phase of running. Biomechanical changes due to shoe midsole resilience seemed to be subject-dependent, while those due to upper structure seemed to be subject-independent.

Parkinson's disease classification using gait analysis via deterministic learning

Gait analysis plays an important role in maintaining the well-being of human mobility and health care, and is a valuable tool for obtaining quantitative information on motor deficits in Parkinson's disease (PD). In this paper, we propose a method to classify (diagnose) patients with PD and healthy control subjects using gait analysis via deterministic learning theory. The classification approach consists of two phases: a training phase and a classification phase. In the training phase, gait characteristics represented by the gait dynamics are derived from the vertical ground reaction forces under the usual and self-selected paces of the subjects. The gait dynamics underlying gait patterns of healthy controls and PD patients are locally accurately approximated by radial basis function (RBF) neural networks. The obtained knowledge of approximated gait dynamics is stored in constant RBF networks. The gait patterns of healthy controls and PD patients constitute a training set. In the classification phase, a bank of dynamical estimators is constructed for all the training gait patterns. Prior knowledge of gait dynamics represented by the constant RBF networks is embedded in the estimators. By comparing the set of estimators with a test gait pattern of a certain PD patient to be classified (diagnosed), a set of classification errors are generated. The average L₁ norms of the errors are taken as the classification measure between the dynamics of the training gait patterns and the dynamics of the test PD gait pattern according to the smallest error principle. When the gait patterns of 93 PD patients and 73 healthy controls are classified with five-fold cross-validation method, the accuracy, sensitivity and specificity of the results are 96.39%, 96.77% and 95.89%, respectively. Based on the results, it may be claimed that the features and the classifiers used in the present study could effectively separate the gait patterns between the groups of PD patients and healthy controls.

Footwear and Foam Surface Alter Gait Initiation of Typical Subjects

Gait initiation is the task commonly used to investigate the anticipatory postural adjustments necessary to begin a new gait cycle from the standing position. In this study, we analyzed whether and how foot-floor interface characteristics influence the gait initiation process. For this purpose, 25 undergraduate students were evaluated while performing a gait initiation task in three experimental conditions: barefoot on a hard surface (barefoot condition), barefoot on a soft surface (foam condition), and shod on a hard surface (shod condition). Two force plates were used to acquire ground reaction forces and moments for each foot separately. A statistical parametric mapping (SPM) analysis was performed in COP time series. We compared the anterior-posterior (AP) and medial-lateral (ML) resultant center of pressure (COP) paths and average velocities, the force peaks under the right and left foot, and the COP integral x force impulse for three different phases: the anticipatory postural adjustment (APA) phase (Phase 1), the swing-foot unloading phase (Phase 2), and the support-foot unloading phase (Phase 3). In Phase 1, significantly smaller ML COP paths and velocities were found for the shod condition compared to the barefoot and foam conditions. Significantly smaller ML COP paths were also found in Phase 2 for the shod condition compared to the barefoot and foam conditions. In Phase 3, increased AP COP velocities were found for the shod condition compared to the barefoot and foam conditions. SPM analysis revealed significant differences for vector COP time series in the shod condition compared to the barefoot and foam conditions. The foam condition limited the impulse-generating capacity of COP shift and produced smaller ML force peaks, resulting in limitations to body-weight transfer from the swing to the support foot. The results suggest that footwear and a soft surface affect COP and impose certain features of gait initiation, especially in the ML direction of Phase 1.

Classification of pallidal oscillations with increasing parkinsonian severity

The firing patterns of neurons in the basal ganglia are known to become more oscillatory and synchronized from healthy to parkinsonian conditions. Similar changes have been observed with local field potentials (LFPs). In this study, we used an unbiased machine learning approach to investigate the utility of pallidal LFPs for discriminating the stages of a progressive parkinsonian model. A feature selection algorithm was used to identify subsets of LFP features that provided the most discriminatory information for severity of parkinsonian motor signs. Prediction errors <20% were achievable using 28 of the possible 206 features tested. For all subjects, a spectral feature within the beta band was chosen through the feature selection algorithm, but a combination of features, including alpha-band power and phase-amplitude coupling, was necessary to achieve minimal prediction errors. There was large variability between the discriminatory features for individual subjects, and testing of classifiers between subjects yielded prediction errors >50%. These results suggest that pallidal oscillations can be predictive biomarkers of parkinsonian severity, but the features are more complex than spectral power in individual frequency bands, such as the beta band. Additionally, the best feature set was subject specific, which highlights the pathophysiological heterogeneity of parkinsonism and the importance of subject specificity when designing closed-loop system controllers dependent on such features.

Long-term evaluation of gait initiation in six Parkinson's disease patients with bilateral subthalamic stimulation

Defined as the transient state between standing and walking, gait initiation is negatively affected in Parkinson's disease (PD), which often results in significant disability. Although deep brain stimulation (DBS) is the most common surgical procedure for PD, the long-term effects of DBS on gait initiation are not well studied. The present study evaluated the long-term effects of subthalamic nucleus (STN) DBS on the preparation phase of gait initiation using principal component (PC) analysis. Six patients with PD who had undergone STN DBS and 24 healthy control subjects were evaluated. PD subjects were assessed 11.3±10.3 (P1) and 78.9±10.6 (P2) months after surgery. PD subjects were tested with STN DBS in two conditions: without medication and with medication. PC analysis was applied separately for the vertical, anterior-posterior and medial-lateral components of ground reaction force (GRF) recorded during gait initiation. Three PC scores were chosen by the scree test for each GRF component and all these PC scores were used for calculating a standard distance between healthy controls and PD subjects. The Friedman test showed a significant difference in standard distance among conditions (P=0.004), with the post-hoc test recognizing differences among P1 conditions and P2 medication-on condition. The eigenvector loading factors pointed to major differences between PD conditions surrounding the maximum amplitude of vertical and anterior-posterior GRF. For the studied sample, all distances increased in the follow-up evaluation (P2) with and without medications, indicating a worsening in gait initiation after seven years.