Self-organized quantification of hormone pulsatility: separating growth hormone secretion in health and disease

摘要

The pulsatile pattern of growth hormone (GH) secretion was assessed over 24 hours in 10 healthy subjects and in 6 patients with a GH producing pituitary tumor (acromegaly) before treatment with the somatosatin analogue octreotide by sampling blood every 10 min. Time series prediction based on a single feedback neural network has recently been demonstrated to separate the secretory dynamics of parathyroid hormone (PTH) in healthy controls from patients with osteoporosis, a severe bone disease. To reveal possible differences of GH secretory dynamics in healthy controls and patients with acromegaly we tested time series prediction based on a single feedforward neural network and a system of multiple neural networks acting in parallel (adaptive mixtures of local experts). Both approaches significantly separated GH dynamics under the various conditions. By performing a self-organized quantification of hormone pulsatility of GH the adaptive mixtures of local experts performed significantly better than the single network approach. It thus may presented a potential tool to characterize alterations of the dynamic regulation in hormonal systems associated with diseased states.