Movement Disorders Moment: Use of 3D Motion Capture for Kinematic Analysis in Movement Disorders

Three-dimensional (3D) motion capture, a digital method of tracking and measuring body movements in space, has applications in various fields, including sports, entertainment, industrial engineering, and clinical practice. The fundamental principle of this technology involves using a precise arrangement of numerous cameras to track and record body motions in 3 dimensions.

State-of-the-art motion capture systems achieve the highest level of accuracy by using reflective markers placed on predetermined anatomic landmarks. Although there is no universally agreed-upon set of kinematic markers, the commonly used ones, such as the Helen Hayes and Cleveland Clinic marker sets, have many similarities and are often referenced. The markers are illuminated by multiple fixed cameras projecting light at specific frequencies, which is then reflected and captured by the cameras. The precise positioning of the cameras enables specialized software to use images from multiple cameras to calculate the exact 3D position of each marker through triangulation. By incorporating kinematic models based on participant-specific measurements and anthropometric reference data, these 3D marker coordinates can be transformed into clinically relevant variables, such as patterns of flexion and extension of individual joints over time. Because of the logistical challenges associated with applying markers and calibrating motion capture systems for individual patients, emerging markerless motion capture technologies use recent advances in computer vision technologies to achieve similar results using synchronized recordings from consumer-grade video cameras. Validation of these markerless approaches against the standard markered systems is ongoing.¹ Data from 3D motion capture can be used in clinical practice to measure movements including joint angles, gait characteristics, and tremor amplitudes and frequency.

It is the authors belief that 3D motion capture holds promise as a valuable tool for neurologists, offering potential benefits by augmenting data captured in validated clinical rating scales, minimizing subjectivity associated with human raters, and enhancing reliability. It allows for continuous measurements of symptom severity, potentially surpassing the precision of ordinal clinical rating scales. This heightened precision may enhance sensitivity in identifying preclinical disease, disease progression, and postintervention changes, and may assist in the diagnostic process (eg, tremor categorization).

Motion Capture

The authors’ center (Emory University School of Medicine, Atlanta, GA) uses a Vicon (Hauppauge, NY) 3D optical motion capture system instrumented with 14 Vicon Vero cameras and 3 FLIR (Wilsonville, OR) Blackfly S BFS-U3-2353C cameras for color video, a raised floor and 2 AMTI (Arlington, VA) HPS400600HF-2K-SYS force plates, Nexus (Austin, TX) v2.15 software, and a Vicon Lock box to integrate analog signals with the system. The motion capture system is capable of triangulating and recording the instantaneous 3D coordinates of each infrared reflective spatial marker attached to the individual’s skin or clothing before motor testing in real-time. Sixty spatial markers are applied to a standardized set of bony landmarks (augmented Helen Hayes full body marker set; Figure 1, A and B). Each marker is an infrared-reflective sphere with a diameter of either 12 mm or 19 mm, for upper- and lower-body placement, respectively. Additional markers are placed on the hands to measure precise movements related to tremor. The markers are attached to the individual with medical-grade, disposable, double-stick adhesive disks. Marker coordinates are computed in real-time from the synchronized video recordings at 120 Hz (frames/second). Each marker is identified and tracked in real-time by matching its relative location within the moving cloud of 60 markers to a topologic template of the individual’s own marker-clad body recorded immediately before motion testing.

Figure 1. Clinical motion capture facility. Our center uses a custom set of 60 retroreflective kinematic markers for most cases. Markers on the hands (arrows) enable tremor measurement. From top to bottom, the markers highlighted are R.Wrist, R.Thumb.M3, and R.Finger3.M3, also shown in Figure 2 (A). After data collection, commercial software triangulates the 3-dimensional coordinates of each kinematic marker in the laboratory space, and assembles a deidentified wire frame or representation of the individual, preserving privacy (B). Our center measures ~650 square feet and is used for both clinical and research applications (C).

The testing area measures 3.0 x 4.6 meters (Figure 1C). Our clinical testing protocol includes multiple upper- and lower-limb tasks performed while seated, standing, and walking. Specific tasks involve assessing tremor while the individual is seated with hands at rest or posture, as well as performing action-dependent tasks, such as finger-to-nose touching. During gait testing, individuals complete at least 5 laps of walking at their usual pace. Turning after laps is recorded as a separate task if the individual can perform this task.

The gathered data include comprehensive information on gait outcome measures (eg, speed, step or stride length, cadence, joint angle variation over time, head or trunk lean, arm swing). In addition, our kinematic output provides data on tremor amplitude and frequency, localized to 16 specific anatomic regions. Over the past decade, there has been a significant increase in the clinical use of motion analysis, particularly in aiding treatment decision-making for conditions such as cerebral palsy and stroke. In this article, we present the emerging use of motion analysis in movement disorders.

Parkinson Disease

Parkinson disease (PD) is a progressive neurodegenerative condition characterized by motor symptoms including bradykinesia, tremor, rigidity, gait impairment, and falls.² Tremor can be a disabling symptom for individuals with PD and is the presenting symptom in approximately 70% to 80% of patients with PD.³ The Movement Disorder Society Unified Parkinson’s Disease Rating Scale motor examination (MDS-UPDRS-III)⁴ is the clinical standard for parkinsonian tremor measurement; however, it is a somewhat subjective assessment, performed with the qualitative judgment of a neurologist. Objective and accurate quantification of tremor is increasingly essential to assess the clinical state as well as to determine the efficacy of therapeutic intervention. To provide a quantitative evaluation of tremor, 3D motion analysis has been used in studies. In a study of individuals with PD with hand tremor, Rajaraman et al⁵ concluded that using 3D motion analysis to analyze tremor provides a robust single quantitative measure of tremor amplitude that is likely to reflect the functional impact of tremor. However, tremor data collection, processing, and quantification algorithms are not standardized across centers. Our laboratory uses a variety of processing pipelines to extract tremor information from kinematic data, similar to workflows in imaging centers. These pipelines vary in their specific methodologies, but the majority take a spectral analysis approach to identify the frequency components of abnormal movements. An example is shown in Figure 2, which illustrates the transformation from raw kinematic data to tremor identification in the frequency domain.

Figure 2. Kinematic analysis of resting tremor in Parkinson disease. Raw traces of the x, y, and z positions of a kinematic marker placed on the third metatarsal of the middle finger of the right hand (R.Finger3.M3) while the individual is sitting quietly are presented. The entire 30-second recording is shown (A). The first 5 seconds of recordings of markers on the middle finger, thumb (R.Thumb.M3), and wrist (R.Wrist) before (left) and after (right) linear filtering to remove low-frequency components are shown. After filtering, low-frequency drift corresponding to voluntary movements is removed (B). The grand mean frequency spectrum (black) of all kinematic markers on the hand (gray) exhibits a clear peak at 5.0 Hz, consistent with Parkinson disease (C).

Gait impairment is a significant symptom and a major cause of morbidity in PD because of the potential for falls. Conventional clinical assessment scores often provide only a superficial evaluation of gait, despite its crucial role in daily activities. For example, the MDS-UPDRS-III only includes 1 measurement to assess gait, scored ordinally from 0 to 4,⁴ and scores critically depend on subjective evaluation. In contrast, 3D motion analysis of gait yields several additional data points, including spatiotemporal measures (ie, step or stride length, step width, velocity, swing) as well as 3D limb kinematics (often expressed in terms of joint range of motion in the sagittal and frontal plane). Figure 3 shows an example from an individual with PD from our center. Each subpanel shows the motion of an individual joint in the sagittal plane averaged over multiple steps and normalized to the gait cycle, which varies from 0% to 100%, corresponding from initial heel strike to subsequent heel strike. By comparing the individual’s joint angle plots (blue) with reference healthy data (gray), reduced arm swing, with a more severe fixed elbow posture on the right side, can be appreciated. However, lower-limb motions are more comparable to those of healthy individuals. At present, 3D gait testing is not widely adopted in neurologic settings, but its extensive utilization in sports medicine and other domains highlights its potential to identify even subtle changes in gait measures after pharmacologic, rehabilitative, or surgical interventions.

Figure 3. Kinematic analysis of gait in Parkinson disease. Images corresponding to the right (A) and left (B) sides of the body are shown. Each subplot shows an individual sagittal plane joint angle versus time, expressed as 0 to 100% of the gait cycle. Comparison of the shoulder and elbow (top 2 rows; blue) versus reference data (gray) reveals frank reduced arm swing (eg, on the right side [A], 3.7 degrees of shoulder motion for the individual versus 21 degrees of motion for the reference data). Note also the increased fixed flexion of the elbow on the right side—an asymmetry common in Parkinson disease. This individual’s lower-limb gait patterns were not remarkably different from those of healthy individuals (HIP_FLEX, KNEE_FLEX, ANK_FLEX).

In a study of individuals with PD using 3D motion analysis,⁶ step length appeared to be the primary determinant of minimum toe clearance, with interventions focused on increasing step length postulated to reduce the risk of trips and falls. Further alterations in interlimb coordination have been detected in PD at the ankle and hip joint during the gait cycle, and alterations in gait pattern during the early swing phase of the gait cycle were noted.⁷ Increased gait parameter variability also has been shown to increase risk of falls in individuals with PD.⁸ A small pilot study also showed that variability of smoothness of measure of the foot and lower-leg segments aided in differentiating individuals with PD from healthy controls.⁹

Freezing of gait (FOG) is a disabling PD symptom that can be unpredictable and challenging to treat. This phenomenon occurs when a person gets “stuck” in place or has a sudden or transient break in the walking motion. Substantial recent progress has been made in using machine-learning approaches to score complex gait impairments (such as FOG) automatically in kinematic and body-worn sensor data. Because of the substantial clinician effort required to annotate the start and stop times of FOG episodes during gait tasks, FOG is typically scored coarsely using only 1 MDS-UPDRS-III entry (item 11) or self-report questionnaires. A machine-learning approach based on convolutional neural networks to score FOG severity automatically in kinematic data with very high concordance with movement-trained clinician ratings has been developed.¹⁰ Ongoing extensions of this work will be used to identify individual FOG episodes. The video below provides an example of a 3D recording of a person with PD demonstrating right hand tremor as well as FOG with normal walk.

In addition, 3D motion capture also can be a useful tool to quantify outcomes after an intervention, such as deep brain stimulation (DBS) therapy. In 1 study of individuals with bilateral subthalamic nucleus (STN)–DBS in the “off” medication state, 3D motion capture identified a significant reduction in tremor amplitude after DBS, demonstrating the therapeutic benefit of this procedure.¹¹ Objective 3D motion capture was used in the same study to measure changes in tremor amplitude associated with visual cues in the DBS “off” and “on” states.¹¹

Furthermore, 3D motion capture also has been used to analyze changes in gait after DBS in PD. As early as 2007, a study demonstrated that both STN-DBS and levodopa therapy increased walking speed and arm and leg swing movement during ambulation.¹² The combination of these 2 treatments augmented each other, yielding additive effects on gait speed as well as upper- and lower-limb range of motion. The possible existence of synergistic effects of STN-DBS and levodopa on gait patterns also was demonstrated in other studies.¹³ This technology also can aid in optimizing DBS therapies; for instance, frequency changes were seen to produce a larger positive effect on gait than voltage changes.¹⁴ With 3D motion capture, clinicians have a unique opportunity to analyze tremor and gait objectively in individuals with PD, providing quantitative data to assist in tracking clinical disease progression and measuring postintervention findings accurately to inform clinical decision-making and improve outcomes.

Essential Tremor and Holmes Tremor

Analysis of tremor using 3D motion capture is not limited to PD, although this is the most common use described in the movement disorder literature. Essential tremor (ET) is the most common cause of action tremor in adults, and it can be disabling, affecting daily activities such as writing, eating, and drinking. ET classically involves the upper limbs and is brought on by arm movements and sustained antigravity posture, but it also may involve other regions of the body, including the head, or the voice. Studies have used 3D motion capture technology to record and specify tremor parameters, such as peak amplitude, frequency, and power spectral density, as well as distribution in different body parts.^15,16 One study evaluated tremor experienced by practitioners while performing endoscopic ear procedures, and described different axial components of the tremor (x, y, or z axis) that may improve with maneuvers such as resting the elbow on the table or manipulating the endoscope by resting it on the ear canal.¹⁷ Another study investigated the possibility of underlying bradykinesia in individuals with ET compared with individuals with PD and healthy controls. Using 3D motion capture to record repeated finger-tapping kinematically, the researchers determined that individuals with ET had slower movements compared with healthy controls and did not have sequence effect (progressive reduction in velocity and amplitude with continued movement), which was present in PD.¹⁸

Holmes tremor is a complex, slow (2-5 Hz), irregular tremor consisting of rest, postural, and action components. It typically arises from neurologic insult to the brainstem, thalamus, or cerebellum and often has other associated neurologic signs. A study using 3D motion capture in age- and sex-matched individuals with Holmes tremor (n=10), PD (n=110), or ET (n=73) showed that both individuals with PD and individuals with ET displayed a statistically significant reduction in tremor amplitude and increase in tremor frequency compared with individuals with Holmes tremor.¹⁹

Dystonia

Dystonia entails sustained contraction of a group of muscles leading to abnormal posturing of a body part. The most common form of dystonia, involving neck muscles (ie, cervical dystonia), has been evaluated using motion analysis to determine abnormal posture and range of neck motion.²⁰ Effects of botulinum toxin therapy in cervical dystonia have been studied using kinematic data such as flexion, extension, and peak angular velocity movements.²¹

Conclusion

Data obtained from 3D motion capture have the potential to contribute greatly to our understanding of movement disorder severity, phenomenology, and response to treatment. With the increasing accessibility of this technology, integrating kinematic data with machine learning holds the potential for earlier, more precise diagnoses, as well as more comprehensive clinical trials.²²