Fig. 3

Decoding Algorithm and illustrative example. Schematic Diagram of the decoder used to regress the CAP event rates to the blood glucose concentration. The means of 20 s, non-overlapping sliding windows over the event rates for the previous 4 min were used as features to the lasso regression model with a regularization factor of 1. The inputs were regressed against a non-time-shifted version of the smoothing spline fit to the measurements of the blood glucose concentration. Leave-one-out cross-validation with 12 folds was used to train 12 separate regression models, and the out-of-sample data was evaluated for each model to reconstruct a piecewise response