Reconstructing the rate of appearance of subcutaneous insulin by deconvolution

In this paper a deconvolution scheme is presented to reconstruct the rate of appearance of subcutaneously injected insulin. Relevant aspects of experiment design are briefly described. Intravenous insulin kinetics are modeled to determine the impulse response of the system. The deconvolution problem is not ill conditioned and is solved using a least-squares method without imposing constraints on the input. An estimate of the error of the reconstructed input is provided. The reliability of the deconvolution scheme is tested by means of an independent validation study. Finally, the different sources of error that affect the method are discussed, and a figure of the global error is derived.