Figure 2.

The features of the artificial neural network used to link smartphone behavior with epileptiform discharges

(A) The overview of the model used to estimate the epileptiform discharge count from a given detector at a particular hour, t (shaded). The smartphone behavioral data spanning a 9 h window were used as the model input (t-4 h to t+4 h). The series of images represent hourly joint interval distributions introduced in Figure 1 with same scales.

(B) A polynomial fit was used to detrend the epileptiform detector counts (see also Figures S2 and S3).

(C) The model input consisted of the joint interval distribution of inter-touch intervals in general and inter-touch intervals based on social apps from the 9-h window. The model output consisted of the counts of the 4 detectors and the counts of the same detectors but detrended for 24 h cycles (labeled as “24 h-d”).