Event Based Modelling

Understanding disease progression pattern is of utmost importance for precise and early diagnosis of neurodegenerative diseases such as Alzheimer’s disease (AD). The disease development is believed to originate from a sequence of causal pathophysiological events which can be attributed to patient's state. However, order of the events remains uncertain. Uncovering the series of the events based on quantifiable biomarkers provide insights into disease development and treatment. The probabilistic generative algorithm called Event-Based Modeling (EBM) to a subset of recognized biomarkers, ranging from the molecular level to neuroimaging features and cognitive test scores to explore sequential AD events.

In order to estimate the most probable temporal sequence of PD risk events, valid progression markers have been collected from ADNI (http://adni.loni.usc.edu). Hypothesizing the optimal operating sequence starts by permuting a set of all possible orders of events in a subgroup of AD biomarkers, namely Aß, Tau, P-tau, Hippocampus volume, Enthorinal volume, Ventricles volume, Mid-temporal volume, MMSE score and wholebrain volume . Subsequently, we assume the distribution of a biomarker in the cohort is a mixture of two normal distributions (healthy and disease). By fitting a Gaussian mixture model to each biomarker, the independent probability of each event to occur and to not occur is calculated. Later, the probability of all possible sequential orders of events is calculated, taking into account the probability of the independent occurrences of each event. Finally, the sequence of the highest probability is considered as the optimal temporal sequence.