Classification Performance Based on Computer Extracted Features

Huo et al. have developed a computer technique for classifying malignant and benign breast masses based on computer-extracted features and have compared the computer accuracy to that of radiologists [58]. In their technique, they used four computer-extracted features (listed in Table 1) that correlate to radiologists' perceptual experience to describe the margin and density of masses. Their comparison of the computer and radiologists was made on a database of 95 mammograms from 65 patients. Of these, 57 mammograms contained a malignant mass and 38 contained a benign mass. On this set of mammograms, the computer achieved an A' of 0.94. The computer's performance was similar to that of an expert mammographer, whose AZ value was 0.91. On the same set of mammograms, five radiologists with some experience in mammography achieved an A' of 0.81. The difference between the A' values of the computer and the five radiologists was statistically significant (p = 0.01).

Was this article helpful?

0 0

Post a comment