Logical classifiers for image analysis in medicine.

A simple, intuitive procedure for the stepwise classification of objects in n-dimensional feature spaces is presented. Each classification step is based on a single feature decision and provides an optimal separation of the objects of only one class from the objects of all other classes. The algorithm of this procedure is implemented on a minicomputer of an image-analysis system. Testing of the classifiers on biomedical data (213 slides of diseased and normal liver tissue) yielded good results requiring few decisions and threshold parameters: 152 correct classifications, 16 incorrect classifications and 45 rejections.