We introduce the notion of an exaggerated image stereotype for some image class of interest, which emphasizes/exaggerates the characteristic patterns in an image class and visualizes what visual information the classication relies on. This is useful for gaining insight into the classi cation and serves for comparison with thebiological models of disease.
We build the exaggerated image stereotypes by optimizing an objective function which consists of a discriminativeterm based on the classi cation accuracy, and a generative term based on the class distribution. Agradient descent method is employed for optimization. We use this idea with Fisher's Linear Discriminant rule,and assume a multivariate normal distribution for samples within a class. The proposed framework is appliedto computed tomography (CT) images of lung tissue with emphysema. The synthesized stereotypes illustratethe exaggerated patterns of lung tissue with emphysema, which is underpinned by three di erent quantitativeevaluation methods.