Galina L. Rogova

Microcalcification texture analysis in a hybrid system for computer-aided mammography

Characterization of microcalcifications with high level of confidence is a very challenging problem since microcalcifications are very small and the difference between benign and malignant clusters is often very subtle. The overall goal of the presented research is to develop a hybrid evidential system for characterization of microcalcifications in order to provide radiologists with a computerized decision aid. The hybrid system intelligently combines a domain knowledge based subsystem with a computer vision subsystem to improve the confidence level of microcalcification characterization. This paper is mainly devoted to the description of the developed computer vision part of the hybrid system. The computer vision subsystem is represented by a hierarchical evidential classifier that computes evidences about the class membership of individual microcalcifications based on their texture and then uses these evidences in a neural network for clusters characterization. The texture of each individual classification is represented by two features: the fractal dimension and a four dimension vector defined by coefficients of the Gabor expansion of a microcalcification image. The results obtained in our experiment prove the feasibility of using this method in the hybrid system.

Information Quality in Information Fusion and Decision Making with Applications to Crisis Management

Designing fusion systems for decision support in complex dynamic situations such as crises requires fusion of a large amount of multimedia and multispectral information coming from geographically distributed sources to produce estimates about objects and gain knowledge of the entire domain of interest. Information to be fused and made sense of includes but is not limited to data obtained from physical sensors, surveillance reports, human intelligence reports, operational information, and information obtained from social medial, opportunistic sensors and traditional open sources (internet, radio, TV, etc.). Successful processing of this information may also demand information sharing and dissemination, and action cooperation of multiple stakeholders. Decision making in such environment calls for designing a fusion-based human–machine system characterizing constant information exchange between all nodes of the processing. The quality of decision making strongly depends on the success of being aware of, and compensating for, insufficient information quality at each step of information exchange. Designing the methods of representing and incorporating information quality into such processing is a relatively new and a rather difficult problem. The chapter discusses major challenges and suggests some approaches to address this problem.

Evidential Approach to Improved Microcalcification Characterization

The overall goal of the research in progress described in this paper is to develop an evidential approach to improved characterization of microcalcifications. Characterization of benign and malignant microcalcifications is very complex and represents a perceptual problem even for an experienced radiologist. Microcalcifications might be very small and the structure of malignant microcalcifications is not much different from that of the benign structures. These perceptual problems result in screening errors which lead either to missed malignant cases or even more often to unnecessary biopsies. All of these factors make the development of an improved automated method for microcalcification characterization a most important and immediate objective.

Reexamining Computational Support for Intelligence Analysis: A Functional Design for a Future Capability

We explore the technological bases for argumentation combined with information fusion techniques to improve intelligence analyses. We review various tools framed by several examples of modern intelligence analyses drawn from different environments. Current tools fail to support computational associations needed for fusion of relations among entities needed for the assembly of an integrated situational picture. Most tools are single-sourced for entity streams, with tools automatically linking analyses between bounded entity-pairs and enabling levels of “data fusion”, but the rigor is limited. Yet these tools often accept the pre-processed extractions from these entities as correct. These tools can identify the intuitive associations among entities, but mostly as if uncertainty did not exist. However, in their attempt to discover relations among entities with little uncertainty and few entity associations, the complexities are left to the human analysts to be resolved. This situation leads to cognitive overloading of the analysts who must manually assemble the selected situational interpretations into a comprehensive narrative. Our goal is automating the integration of complex hypotheses. We review the literature of computational support for argumentation and, for an integrated functional design, as part of a combined approach, we nominate a unique, belief- and story-based subsystem designed to support hybrid argumentation. To deal with the largely textual data foundation of these intelligence analyses, we describe how a previously, author-developed, ‘hard plus soft’ information fusion system (combining sensor/hard and textual/soft information) could be integrated into a functional design. We combine these two unique capabilities into a scheme that arguably overcomes many of the deficiencies we cite to provide considerable improvement in efficiency and effectiveness for intelligence analyses.

Decision-making involving low probability high consequence events under risk and uncertainty

Research in progress described in this paper addresses the problem of decision making in situations involving low probability high consequence events. The traditional Expected Utility Model (EU) has significant limitations in such circumstances as documented in multiple research results. The models discussed in this paper is an adaptation of the Multiple Quantile Model (MQT) representing a rational decision support scheme suited to regular as well as low probability high consequence events to the complex dynamic scenarios, in which decision making has to be based on highly uncertain, often unreliable heterogeneous data and information. The core of this scheme is a combination of the Multiple Quantile Theory with the Transferable Belief Model (TBM) and Anytime Decision making. An example of this approach with numeric simulations is given and the directions of future work are outlined.

Belief-Based Argumentation in Intelligence Analysis and Decision Making

This paper asserts that a multi-perspective viewpoint must be taken in the design of a computational system support capability for decision-making. We offer views from a Decision-Science slant, a Systemic Architectural view, and the need for technological support to realize improvements in analytical rigor. We have been researching and evolving the design of an analysis tool framework exploiting the hybrid concepts of a Belief-based Argumentation and Story-based subsystem. The notion of rigor, defined as a quality measure on the reasoning/analysis process, is one overarching principle of our approach, driven by the need for the associated analysis/decision-support product quality that complex modern problems demand. Our approach to the design of a mixed-initiative analysis tool is highly multidisciplinary and has taken account of an exhaustive review of the relevant literature along each viewpoint.

Crisis Management and Context

Successful management of critical situations created by major natural and man-made activities requires monitoring, recognizing, fusing, and making sense of these activities in order to support decision makers in either preventing a crisis or acting effectively to mitigate its adverse impact. Context plays an important role in crisis management since it provides decision makers with important knowledge about current situations and situation dynamics in relation to their goals, functions, and information needs, to enable them to appropriately adapt their decisions and actions. Efficient context exploitation for crisis management requires a clear understanding of what context is, how to represent it and use it. The chapter provides a brief discussion of the key issues of the problem of context definition, representation, discovery, and utilization in crisis management.

Formalization of “Context” for Information Fusion

Context exploitation can provide benefits for information fusion by establishing expectations of world states, explaining received data, and resolving ambiguous interpretations; thereby improving process efficiency, reliability, and trustworthiness of the fusion product. While everybody recognizes the importance of considering context in inferencing, designers of information fusion processes only recently have begun to incorporate context explicitly into fusion processes. Effective context exploitation requires a clear understanding of what context is, how to represent it in a formal way, and how to use it for particular information fusion applications. Although these problems are similar to the ones discussed by researchers in many other fields, consideration of context in designing information fusion systems also poses additional challenges such as understanding the relationships between situations and context, utilizing context for understanding and fusion of natural language data, context dynamics, context recognition, and contextual reasoning under the uncertainty inherent in fusion problems. This chapter provides a brief discussion on possible ways of confronting these challenges while designing information fusion systems.

System-Level Use of Contextual Information

A system that exploits information—e.g. to support decision making—can use contextual information both in providing expectations and in resolving uncertain inferences. In the latter case, contextual reasoning involves inferring desired information (values of “problem variables”) on the basis of other available information (“context variables”). Relevant contexts are often not self-evident, but must be discovered or selected as a means to problem-solving. Therefore, context exploitation involves (a) predicting the value of contextual information to meet information needs; (b) selecting information types and sources expected to provide information useful in meeting those needs; (c) determining the relevance and quality of acquired information; and (d) applying selected information to a problem at hand. Fusion of contextual information can improve the quality of inferences, but involves concerns about the quality of the contextual information. The availability and quality of predictive models dictate the ways in which contextual information can be used. Many applications are benefitted by inference systems that adaptively discover and exploit context and refine such models to meet evolving information states and information needs.