Medical applications of image understanding

Medical imaging problems often deal with geometrically intricate objects, such as the surface of the cortex or the bronchial structure of the lungs. Medical problems also deal with flexible, deformable objects, as for example the nonrigid motion of the beating heart. Despite the apparent differences in underlying bases, developments over the past few decades in computer vision, especially through ARPA´s Image Understanding Program, are beginning to revolutionize the use of medical imagery: in surgery, in diagnosis, and in therapy evaluation. This applicability of IU techniques to medical problems holds for many of the central problems in traditional IU: extracting key features from the imagery, registering data sets, predicting images of object models from arbitrary viewpoints, and fitting parameterized surface models to data. By tailoring IU techniques designed for traditional vision to the special circumstances of medical imagery, systems are emerging that support effective use of medical image data. To demonstrate the range of roles that IU methods play in medical image utilization, the author describes an end-to-end system for image guided surgery. This system directly builds on a wide range of IU methods to provide surgeons with visualization and guidance during surgical procedures