Interactive image cases for medical problem solving

The development of an intelligent assistance agent for diagnostic radiology faces a number of challenges: e.g., how to gain access to the knowledge used in such visual problem solving tasks, how to operationalize it for further processing, and how to present it in the most useful manner for diagnosis. Our previous work in this domain placed an emphasis on the first two issues, and has shown that both perception and problem solving need to be supported in these types of tasks. However, the interface to the underlying intelligence in our first prototypes did not reflect how medical practitioners physically interact with images in the context of their work. The paper describes current work in progress on an image-centered approach to the agent's interface design. This approach will allow users to interact dynamically with the image, and to store these interactions for future retrieval.