Reliability of visual observation to classify continuous wavelet transforms of signals of extremity movements

This dataset accompanies the research study titled "Reliability of visual observation to classify continuous wavelet transforms of signals of extremity movements." The primary aim of this work is to investigate the feasibility and consistency of human visual assessment of accelerometer-derived time-frequency signals in evaluating motor abnormalities, particularly in individuals diagnosed with Parkinson’s disease (PD) compared to healthy individuals with typical development (TD).

The dataset comprises continuous wavelet transform (CWT) images that were generated from raw accelerometer signals collected during a series of standardized motor tasks. These tasks, adapted from the Movement Disorder Society-sponsored revision of the Unified Parkinson’s Disease Rating Scale (MDS-UPDRS), included repetitive movements such as finger tapping, toe tapping, pronation-supination, hand movements, and leg agility. Participants were equipped with low-cost wearable sensors, taped directly to their extremities, as described by McKay et al. (2019). These sensors captured high-resolution, continuous movement data during in-person clinical assessments.

The raw accelerometer signals were processed and converted into continuous wavelet transforms, creating two-dimensional time-frequency images that visualize the movement patterns. These CWT images served as the basis for a blinded observational analysis. A group of 31 trained raters—who were blinded to all participant-specific information, including age, sex, clinical diagnosis, and which limb the signal was collected from—systematically scored each image for signs of motor dysfunction. Specifically, raters evaluated each image for abnormalities across three parameters: amplitude, frequency, and interruptions in the movement signal.

This methodology builds upon and extends previous research by Hernandez et al. (2022) and Ziegelman et al. (2023), who demonstrated that visual interpretation of wavelet-transformed sensor data could be a viable remote assessment tool for movement disorders. Statistical validation of inter-rater reliability was carried out using Cronbach’s alpha and intraclass correlation coefficients (ICC), with analysis performed using IBM SPSS Statistics version 30.0.0. The high ICC values observed in this study support the reproducibility and reliability of this visual classification approach.

Moreover, this work reinforces and expands upon findings by Elshourbagy et al. (2023), who previously assessed the feasibility of using similar low-cost, sensor-based technologies for remote motor evaluations. It also complements the clinically validated MDS-UPDRS framework outlined by Goetz et al. (2008), by offering a more accessible, potentially scalable method for motor assessment using visual data derived from wearable devices.