A method for generating a synthesized medical image by receiving a normal image includes generating first data based on a random selection, generating second data, and, based at least in part on the first and second data, modifying the normal image to form the synthesized medical image. Modifying the normal image comprises combining the first data and the second data. The first data characterizes an image that represents a lesion and the second data characterizes a transformation of that image as well as a location of the lesion.