Figure 1.

(A) (Left) two-dimensional embedding (t-Distributed Stochastic Neighbor Embedding) (50) of acceleration data points from nine healthy subjects performing several daily activities. Data were collected with one inertial measurement unit (IMU) placed over the pelvis with a sample rate of 120 Hz. Each point corresponds to a discretized behavior block obtained with a change-point detection algorithm. Colors correspond to clusters obtained by affinity propagation (51). A subset of clusters was labeled according to the dominant feature (standing, walking, sitting, etc.). (B) (Right) shows an example of the same methodology now applied to six patients diagnosed with idiopathic Parkinson’s disease (PD) and a Hoehn and Yahr stage less than or equal to 2. Movement data were collected with one IMU placed over the pelvis (sample rate 120 Hz) while patients walked a 10-m corridor, first in their OFF state (“x”) and later in their best ON state defined by the Movement Disorder Society Unified Parkinson’s disease rating scale (“o”). The gray dots represent all the recorded blocks of behavior from the six PD patients and the color dots the groups related to one patient. Clusters represented in both left and right images are from two independent experiments, thus, the resultant embeddings are not comparable.