Given the recognized and widely accepted importance of mammography in early identification of breast cancer in general, and in screening, in particular, computer-aided diagnosis (CAD) schemes for mammography have been developed and investigated extensively in recent years. The diagnostic performance of some CAD systems is at the point where clinical utility has been demonstrated. However, despite significant efforts and major improvements, the performance level of CAD schemes, particularly as related to masses, remains less than optimal. In addition, it is becoming increasingly important to explore ways not only to optimize these schemes, but also to assess and compare their performance and robustness when used in the clinical environment. As we increase efforts to detect abnormalities at an earlier stage, this field is likely to progress incrementally in several important areas. In this project, we propose to continue to conduct a variety of studies to improve and test the performance and robustness of several of our own CAD schemes. We will also assess a number of issues associated with system's performance when multiple images, such as multiple views or sequences of images, are available and are used to optimize the results of a diagnostic examination. Many of the methods we will develop and the techniques we propose to use in order to optimize performance and assure robustness of these schemes should be applicable to other investigations in this and related fields. Last, there is no direct comparison of schemes developed by different groups (and companies) that will enable a better understanding of some of the advantages and limitations of different approaches. We propose to do so twice during the project on a large number of biopsy-proven cases.
Showing the most recent 10 out of 43 publications
