Abstract
To investigate the influence of movement trajectory segmentation on classification performance of outcome measures for sensorimotor control evaluation of neck pain-related conditions.
Outcome measure improvement analysis. Subjects performed the Butterfly Test (NeckCare System, NeckCare) wearing a head-mounted movement sensor and attempted to accurately track a moving target on a computer screen by controlling a cursor via the sensor. The target moved unpredictably along 3 predetermined trajectories. An offline change point detection method was used to segment each trajectory into intervals that each represented a unique movement profile. Segmentation was performed based on average movement accuracy (cursor deviation from target) of all participants. Due to the underrepresentation of subjects with neck pain, an upsampling method was applied to balance the dataset. The dataset was normalized so that parameters had a mean of 0 and a standard deviation of 1. A multilayer perceptron model was used for classification. Model hyperparameters were selected using hyperparameter tuning, and the model was trained using 5-fold cross-validation.
Data were collected in a laboratory setting (healthy) and at an outpatient clinic (patients).
Eighty-six healthy subjects (48 females, 38 males) and 31 subjects with neck pain of various origins (19 women, 12 men, visual analog scale score: 3.5±2.5) participated in the study. Healthy subjects were recruited using convenience sampling. For the neck pain group, individuals seeking physical therapy for neck pain were recruited from 2 outpatient clinics.
Not applicable.
The Butterfly Test defines 4 parameters that quantify movement performance and are used to detect abnormal movement patterns indicative of impaired sensorimotor control. The parameters included the average deviation between the subject-controlled cursor and the target and the relative time spent on, behind, or ahead of the target. The parameters were determined for each segmented interval.
The data segmentation process identified 12 distinct movement profiles from the 3 movement trajectories. Therefore, parameters in the classification feature set increased from 12 (4 per trajectory) to 48 (12-24 per trajectory). Data segmentation resulted in a substantial improvement in classification accuracy, sensitivity (true positive rate), and specificity (true negative rate). Accuracy increased from 81.6% to 94.9%, sensitivity from 80.0% to 90.3%, and specificity from 82.9% to 96.7%.
Segmenting movement trajectory data into distinct movement profiles enhanced the classification performance of the Butterfly Test for patients with heterogeneous neck pain-related conditions. Accurate detection of abnormal neck movement patterns may support clinical decision making and inform targeted interventions.
Jona Olafsdottir owns shares in NeckCare. The other authors have nothing to disclose.