Uncertain decision-making schemes for knowledge based anatomical landmark localization (K-BALL)

We have proposed a method for localization of anatomical landmarks comprised of two steps: 1) information extraction, and 2) information analysis. The focus of this paper is on the second step during which a set of points, found in the first step, is evaluated. This step utilizes a set of rules and eventually generates a confidence factor (CNF) for each set of points that indicates as to what degree they are good representatives of the "landmarks of interest". We also propose two alternative decision-making schemes: 1) Bayesian networks, and 2) possibilistic inference methods. The latter design is expected to outperform the other two as it utilizes both desired and undesired information. Our simulation study resulted in very similar performance for the rule-based and Bayesian schemes. The overall success rate (average of sensitivity and specificity) of the entire proposed method in localization of the hippocampus on MRI images was 83.3% with an accuracy of 99.2% (using rule-based decision-making scheme).