INTRODUCTION In 1995, an estimated 183,400 women in the United States will be newly diagnosed with breast cancer, and 46,240 will die of the disease [1]. Screening mammography effectively detects early breast cancers and can increase the likelihood of cure and long-term survival [2]. Differentiating between benign and malignant mammographic findings, however, is difficult. Only 15%-30% of biopsies performed on nonpalpable but mammographically suspicious lesions prove malignant [3]. Automated classification of mammographic findings using discriminant analysis and artificial neural networks has indicated the potential usefulness of computeraided diagnosis [4,5]. We explored the use of Bayesian networks as a * Section of Information and Decision Sciences, Department of Radiology, Medical College of Wisconsin, Milwaukee, Wisconsin 53226-0509; and Artificial Intelligence Laboratory, Department of Electrical Engineering and Computer Science, University of Wisconsin-Mil

. This paper investigates the power of first-order probabilistic logic (FOPL) as a representation language for complex dynamic situations. We introduce a sublanguage of FOPL and use it to provide a first-order version of dynamic belief networks.We show that this language is expressive enough to enable reasoning over time and to allow procedural representations of conditional probability tables. In particular, we define decision tree representations of conditional probability tables that can be used to decrease the size of the created belief networks. We provide an inference algorithm for our sublanguage using the paradigm of knowledge-based model construction. Given a FOPL knowledge base and a particular situation, our algorithm constructs a propositional dynamic belief network, which can be solved using standard belief network inference algorithms. In contrast to common dynamic belief networks, the structure of our networks is more flexible and better adapted to the given ...