Objective: To evaluate an approach of visual contact-free tremor detection using infrared video capabilities of the Microsoft Kinect™ sensor.

Background: Tremor is one of the most frequent hyperkinetic movement disorders and its phenomenology is usually described by location, occurrence at rest or in action, permanence, amplitude and frequency. Whereas clinical severity ratings are based on the features of permanence and amplitude, tremor frequency is an important feature for differential diagnosis. Current standards of instrumental tremor analysis require multi-channel recordings of electromyogram (EMG) and accelerometry. This requires marker placement and limits the applicability to stationary positions and to preselected body regions.

Methods: 17 Patients (age: 49-83, m:5 w:12) with focal tremulous dystonia were simultaneously recorded with EMG, 2 biaxial accelerometers (ACC) attached to the frontal and parietal forehead (Nicolet EDX system) and a Microsoft Kinect V2.0 sensor. Each patient performed up to 5 different postural tasks (including sitting with: eyes open, eyes closed, eyes closed + mental distraction) resulting in 74 measurements that were available for analysis. Face detection (Viola-Jones and Kanade-Lucas-Tomasi algorithms) was applied to Kinect infrared recordings and movements of resulting feature points were used to calculate main movement frequencies, which were compared to main frequencies derived from ACC and EMG. Tremor was assumed for main frequencies between 1Hz and 10Hz and a power magnitude above 2.

Results: Visual inspection by a clinician blinded to results of tremor analysis detected tremors in 46 out of 74 recordings (TETRAS tremor ratings 1.6 ± 0.08). Kinect based tremor detection showed best sensitivity (Kinect: 100%, ACC: 34.8%, EMG: 60.9%) but also worst specificity (Kinect: 39.4%, ACC:97.4%, EMG:73.7%). Differences in detected main frequencies were present (EMG: f = 4.1Hz, ACC: f=4.5Hz, Kinect: f=3.3Hz).

Conclusions: (1) Kinect is capable to detect clinically apparent head tremor in patients with focal tremulous dystonia; (2) sensitivity of Kinect to detect clinically overt tremor is superior to ACC and EMG and (3) rate of false positive tremor detection is higher in Kinect compared to conventional tremor analysis (TA). (4) non-detection of tremor may be related to irregular, low-amplitude and inconstant tremors for all methods studied.

« Back to 2018 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/evaluation-of-visual-perceptive-computing-for-tremor-analysis/