EMG and IMU data for sitting knee extension/bending movements

Highly accurate and lightweight automated movements quality assessment is essential for home rehabilitation patients. We propose a method for the assessment and quantification of movement quality based on the differential feature segments, the objective  is to emulate the expert evaluations of physicians as closely as possible with minimal data features. Employing the Gaussian mixture model (GMM) to divide continuous trend time-series data into fragment features, defined as feature segments. Calculating the log-likelihood of sample movement feature segments to their corresponding standard movement feature segments, then fits these calculations into Fuzzy comprehensive evaluation (FCE) results to quantify assessment scores. We used the seated knee extension/flexion movement to validate, collecting data from inertial measurement units (IMU) and electromyography (EMG) sensors. Four boosting algorithms were tested in the data analysis experiments, the results demonstrated that using as few as two sensors and the LightGBM algorithm could emulate the physician's FCE estimate with a determination coefficient of 0.84. Compared to Dynamic time warping (DTW) and traditional GMM approaches, the proposed method based on segmented feature segments yielded superior GMM quantified regression scores, showing higher correlation with FCE outcomes. This method could maximally utilizes the information in time-series data to closely emulate physician evaluations with a minimal amount of data features.