Status Report

  • Hybrid Pyramid/Neural Network Vision System

    January 17, 2017

    An artificial problem was constructed and described in the last report. During this quarter, another, more difficult, artificial problem was constructed. As before, the objects to be found and some potential false positives each have component patterns. Each positive has two different component patterns chosen from three types. In addition, the rectangles which form the objects and the sub-patterns all have artificial shadows added (Figure 1). The angle of the shadow is randomly-chosen. The potential false positive objects either have two component patterns of the same type, or only one pattern of some type, and one or both of the component patterns may not have a shadow. Thus the pattern tree must detect all three types of components with shadows to be able to detect a positive. As in the previous simpler problem, the positives and potential false-positives are 18-by-11 pixel rectangles, but now their brightness is randomly-chosen between 136 to 247. The sub-patterns have a brightness that is independently chosen at random from the same range, but both patterns (if there are two in an object) have the same brightness. The component patterns are three-by-three x, +, and square patterns.

  • Application of Information Theory to Improve Computer-Aided Diagnosis

    January 17, 2017

    Mammographic Computer-Aided Diagnosis (CAD) systems are an approach for low-cost double reading. Though results to date have been promising, current systems often suffer from unacceptably high false positive rates. Improved methods are needed for optimally setting the system parameters, particularly in the case of statistical models that are common elements of most CAD systems. In this research project we developed a framework for building hierarchical pattern recognizers for CAD based on information theoretic criteria, e.g., the minimum description length (MDL). As part of this framework, we developed a hierarchical image probability (HIP) model. HIP models are well-suited to information theoretic methods since they are generative. We developed architecture search algorithms based on information theory, and applied these to mammographic CAD. The resulting mass detection algorithm, for example, reduced the false positive rate of a CAD system by 30% with no loss of sensitivity. We showed that the criteria reliably correlate with performance on new data. The framework allows many other applications not possible with most pattern recognition algorithms, including rejection of novel examples that can’t be reliably classified, synthesis of artificial images to investigate the structure learned by the model, and compression, which is as good as JPEG.