Application of Self-Organization Maps to the Biomedical Images Classification

Abstract

A diagnostic system was presented that employs multifractal analysis combined with self-organization maps approach, for the discrimination normal cells from malignant. The input to the system consists of images of routine processed cervical smears stained by Papanicolaou technique. The analysis of the images provided a data set of cell features. The neural network classifier, an efficient pattern recognition approach, was used to classify normal and malignant cells based on the extracted multifractal features. The application of self-organization map yielded high rates of correct classification at both the cell level and the patient level. These results indicate that the use of intelligent computational techniques along with multifractal features may offer very useful information about the potential of malignancy of cervical cells.