To design, optimize, verify and validate smartphone-instrumented tests, and their sensor-based algorithms, for extracting reliable digital biomarkers (DB) in neurological patients.

Modern smartphones can provide an accurate and reliable assessment of neurological functions. However, the development of smartphone-instrumented tests and their algorithms is a challenging task that needs to overcome technological limitations, particularly in relation to the quality of the sensors and the derivation of DB insensitive to the orientation and position of the devices.

100 healthy participants and 59 patients with Multiple Sclerosis (PwMS) were included. They performed eight smartphone-instrumented tests (dreaMS, NCT05009160): Screen-to-nose, timed-up-and-go, 30-second-walk, chair-rising, stair-climbing, tandem-walking, standing-balance, and arm-balance. The participants were also equipped with a motion capture system (Xsens-MVN). The methodology comprised the technical verification of the ground-truth datasets, and the analytical and clinical validation of DB.

The final ground-truth datasets comprised 3015/3272 test executions verified by two raters. Smartphone signals showed a strong Pearson correlation (0.997 median correlation - [range 0.889-1.0]) with the motion-signals of the XSens-MVN system. A total of 66 biomarkers showed good levels of agreement with the ground-truth measurements; Intra Class Correlations (ICC) > 0.6 [range 0.601-0.999]. Also, 347 biomarkers were reliable on the population of healthy participants; ICC > 0.6 [range 0.601-0.966] or median coefficient of variations (mCV) < 20% [range 1%-19.7%].