Head tremor in cervical dystonia: Quantifying severity with computer vision

Abstract

Background: Head tremor (HT) is a common feature of cervical dystonia (CD), usually quantified by subjective observation. Technological developments offer alternatives for measuring HT severity that are objective and amenable to automation.

Objectives: Our objectives were to develop CMOR (Computational Motor Objective Rater; a computer vision-based software system) to quantify oscillatory and directional aspects of HT from video recordings during a clinical examination and to test its convergent validity with clinical rating scales.

Methods: For 93 participants with isolated CD and HT enrolled by the Dystonia Coalition, we analyzed video recordings from an examination segment in which participants were instructed to let their head drift to its most comfortable dystonic position. We evaluated peak power, frequency, and directional dominance, and used Spearman's correlation to measure the agreement between CMOR and clinical ratings.

Results: Power averaged 0.90 (SD 1.80) deg²/Hz, and peak frequency 1.95 (SD 0.94) Hz. The dominant HT axis was pitch (antero/retrocollis) for 50%, roll (laterocollis) for 6%, and yaw (torticollis) for 44% of participants. One-sided t-tests showed substantial contributions from the secondary (t = 18.17, p < 0.0001) and tertiary (t = 12.89, p < 0.0001) HT axes. CMOR's HT severity measure positively correlated with the HT item on the Toronto Western Spasmodic Torticollis Rating Scale-2 (Spearman's rho = 0.54, p < 0.001).

Conclusions: We demonstrate a new objective method to measure HT severity that requires only conventional video recordings, quantifies the complexities of HT in CD, and exhibits convergent validity with clinical severity ratings.

Keywords: Computer vision; Head tremor; Severity rating; TWSTRS; Video.