An intelligent system for the diagnosis of complex images

An intelligent system suitable to perform a computer aided diagnosis of complex images should have a knowledge base containing all information related both to the images to be interpreted and to their symbolic description. In this paper, a system able to classify unknown medical digital images into four classes is proposed (searched pathology recognized, searched pathology absent, different pathology from the searched one recognized, unknown pathology). A main component of this system is a knowledge base that, startling from information deduced from sample images, can be processed to create synthetic reference models that, in turn, permit the interpretation of real scenes. The system has been tested on digitized plain film of the thorax, in order to perform a computer-aided diagnosis of pneumothorax cases.