Rule-based decision-making framework for knowledge-based anatomical landmark localization (K-BALL)

K-BALL is a general method for localization of anatomical phenomena of the same origin with natural discrepancies distributed over a reference space, e.g., human brain anatomical structures. In this paper, we focus on information analysis step (2nd step) of K-BALL during which landmarks extracted in its first step are evaluated. We provide a framework in which rules are automatically generated based on estimated and derived models. We show that the rules based on the derived models can improve the overall success rate of K-BALL. Each rule evaluates the extracted points by producing an intermediate confidence factor (ICNF). A total confidence factor is calculated using ICNF´s to facilitate the acceptance or rejection of a set of points as landmarks of interest. Using the rules merely based on the estimated models, simulation study produced an overall success rate of 91.8%. Using the rules based on both of the estimated and derived models, this rate increased to 92.5%