Abstract

The objective of this paper is to explore the implementation of a Bayesian Belief Network for an automated breast cancer detection support tool. It is intuitive that Bayesian networks are employed as one viable option for computer-aided detection by representing the relationships between diagnoses, physical findings, laboratory test results and imaging study findings. This work brings important entities such as Radiologists, Image Processing Scientists, Data Base Specialists and Applied Mathematicians on a common platform. A brief background concerning causal networks, probability theory and Bayesian networks is given; available computational tools and platforms are described. It is explained that, by exploiting conditional independencies entailed by influence chains, it is possible to represent a large instance in a Bayesian network using little space, and it is often possible to perform probabilistic inference among the features in an acceptable amount of time. The next steps towards realizing a Bayesian Belief Network Implementation are described. Bayesian networks have an unparallel advantage of being able to exploit the explicit structure of the domain model to derive a graphical representation for learning. The encoding of independencies in the network topology admits the design of efficient procedures for performing computations over the network. For the application of computer-aided detection in mammography, the researchers intend to design an interface between the project's Bayesian network learning algorithm and the radiologists, so that the radiologists can have interaction with the system by labeling only a small number of informative images presented by the active learning algorithm.

Key words: Bayesian networks, breast cancer, computer aided decision support.