Deep Learning and Statistical longitudinal medication studies of attention-deficit/hyperactivity disorder of (ADHD) participants

5-8% of children in the US are diagnosed with ADHD. They show lack of attention, often with significant kinematic motor skill impairments. We pursue the hypothesis that important cognitive information is contained on humans’ movements when looked at millisecond time scales, away from naked eye detection. We used high-definition Bluetooth sensors to measure the kinematic changes over several hours on ADHD participants that carryed out hand reaching experiments after taking their medication. We first analyzed the raw kinematic data using Deep Learning (DL) techniques. The DL could classify an unseen time-series of kinematic data based on how long it has been since medication. Next, we filtered the electronic sensors noise finding kinematic millisecond random fluctuations, leading to histograms from the nearest neighbor magnitude fluctuations for each ADHD participant. We calculated the Fano Factor and Shannon entropy biometrics as a function of time showing the motor medication effects. One participant starts by showing the baseline random histogram, before medication. After a few hours there is an intermediate histogram that looks like the one for typical development. But thereafter it goes back to the more random histogram when the effect of medication stops. We will test other participants to assess the generality of these results. They may be connected to a participants’ treatment response, as well as the effectiveness of the therapies. This information maybe of importance to know by providers.