PROVIDING CONTEXT FOR TUMOR RECOGNITION USING THE WRAPPER FRAMEWORK

In recent decades, imaging became a very powerful tool, offering new possibilities for medical diagnostic investigations. Dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) has become an important source of information to improve cancer diagnosis; however, evaluation of an image is a challenging task, requiring automated means of assisting the human expert. Automatic detection implies there is a classification process to identify objects requiring further human analysis. Traditionally classification in computer vision assumes that the object of interest can be successfully segmented from the background, however this is extremely difficult without contextual knowledge about the object of interest. We propose providing contextual information both via the shape of the objects of interest and the temporal kinetic signal. The approach is demonstrated on the application of using of machine vision to classify breast disorders into four classes: (i) normal, (ii) benign, (ii) suspicious, and (iv) malignant. Preliminary results show we can achieve a 92% automatic recognition rate.