LEARNING OF UNCERTAIN CLASSIFICATION RULES IN BIOMEDICAL IMAGES - THE CASE OF COLPOSCOPIC IMAGES

Knowledge acquisition is always a critical step in the development of a knowledge-based computing system. In the particular area of the interpretation of biomedical images, the assignment of meanings to image patterns is based on obscure and intrinsically vague criteria which are difficult to assess and transform into a suitable machine representation. Automatic learning techniques may be a promising tool in addressing this problem. The paper illustrates a methodological procedure based on fuzzy set theory and using fuzzy logic for the automatic learning of classification rules for biomedical image interpretation systems. It also provides a detailed description of the application of the procedure in the development of a system for the automatic detection of preneoplastic and neoplastic lesions in colposcopic images. Plans to employ the system contemplate its use in educational applications, in diagnostic review for research purposes, and as an online support in clinical practice.