Detection of Dystonia in Cerebral Palsy Using Dyskinesia Impairment Scale and Machine Learning

The severity of dystonia must be accurately and consistently measured in order to identify, evaluate, monitor, and improve therapy. Currently, it is common practice to assess dystonia in kids and teenagers visually, either in the hospital environment or by grading video clips captured in other place using standardized scales. Either of these methods need a lot of clinical specialists' time and effort and are lacking in objectivity. As choreoathetosis and dystonia are known to change over time and can get worse owing to exhaustion, pain, stress, or emotional causes, all of which are frequent occurrences in a clinical context, they only give a fleeting glimpse of the severity of dyskinetic movements. The goal of current research is to diagnose the viability of using smartphone-connected inertial sensors and machine learning for measuring the severity of dystonia at home. In nonclinical settings, video and sensor data were collected from 20 patients in active and resting situations. Using a 0–4 scale and conforming to the amplitude specification of the Dyskinesia Impairment Scale (DIS), three clinicians evaluated the extremity dystonia. The sensor data and clinical results were integrated to train various machine learning models through cross-validation. The average F1 scores from independent test datasets (0.99 0.98 for lower limbs and 0.97 0.95 for upper body parts) indicate the viability of automatically identifying dystonia using models that were independently trained. These forecasts could be useful supplements to current dyskinetic CP tests, providing regular, unbiased assessments in practical contexts that could enhance clinical management. A generalized model trained on data from several participants results in lower F1 scores (0.95 for the lower limbs and 0.54 for the upper body parts) because of the limited training data and subject variability. However, the generalized model demonstrates a commendable ability to differentiate between high and low scores.