ABSTRACT
This chapter deals with evaluation criteria for software designed to detect abnormalities in medical images. When computer algorithms are designed for applications such as this, strict evaluation criteria are crucial for a number of reasons. The most obvious reason is the need to assess expected system performance. This may be required to gain widespread acceptance or to establish the state of the art. In addition, some means of objective and fair comparison are needed to help determine the relative merit of competing algorithms. A rigorous evaluation also facilitates the development of a robust system. By knowing specifically which situations cause an algorithm to fail, we can begin to isolate and address the causes of the failure. This leads to further development in an effort to correct the problem, handle the error, or enable the algorithm to recognize when failure occurs.
