Plaque Distribution Patterns

In existing studies, all plaques within the coronary artery tree have been regarded independently. A possible correlation among the localization of plaques was not regarded. For the first time, we therefore researched spatial distribution patterns of atherosclerosis in the coronary artery tree by using the frequent itemset mining algorithm.

Frequent itemset mining - originally introduced by Agrawal et al. - is used to determine rules that describe spatial correlations between plaques. The introduction of frequent itemset mining was driven by market basket analysis. In such an analysis, the items in customers' baskets are analyzed to find sets of products that are frequently bought together. After the observation of a number of baskets, so-called association or prediction rules can be derived to express the probability that - given a set of items (itemset) in a basket - a certain other item is also present. These prediction rules represent statistical correlations which can be used to research the dependencies among items and to predict the content of new baskets if only some items are given.

For the research of spatial distribution patterns of atherosclerosis, we adopted the market basket analysis by frequent itemset mining to our special purpose. Hence, a patient corresponds to a basket and existing plaques are described as items. Multiple items in form of an itemset tree are defined and assigned to each plaque. Afterwards, these items are analyzed with frequent itemset mining and prediction rules are extracted to express the observed distribution patterns. We now uses these prediction rules to propose a guided review for improved plaque detection.

Guided review by frequent itemset mining has been successfully evaluated in several studies and can be tested as a web application.