Validation of Medical Image Analysis Techniques

Bowyer, Kevin W.

doi:10.1117/3.831079.ch10

Book: Handbook of Medical Imaging, Volume 2. Medical Image Processing and Analysis

Editor(s): J. Michael Fitzpatrick; Milan Sonka

Author(s): Kevin Bowyer

Published: 2000

https://doi.org/10.1117/3.831079.ch10

Abstract

This chapter focuses on the problem of measuring the performance of medical image processing and analysis techniques. There are two important, but quite different, possible definitions of the term "performance" in this context. One definition of performance is the time taken to execute an algorithm on a computer. On this dimension, an algorithm that requires less computation time is better. The other definition of performance relates to how frequently an algorithm results in a correct decision. On this dimension, a better algorithm is one that gives a correct decision more frequently. This chapter is concerned with measuring the performance of algorithms in this second sense. As one example, we want to be able to make meaningful statements of the form - "Algorithm A performs better than algorithm B at detecting signs of cancer in medical images" or "Algorithm A performs better than algorithm B at registering CT and MRI images of the brain." It is only when algorithms have similar performance on this dimension that their relative computation time becomes of interest.