Bildsegmentation zur Untersuchung von Streulichtbildern bei der laseroptischen Diagnose von rheumatoider Arthritis

Abstract

Optical imaging in biomedicine is governed by the light absorption and scattering interaction on microscopic and macroscopic constituents in the medium. Therefore, light scattering characteristics of human tissue correlates with the stage of some diseases. In the near infrared range the scattering event with the coefficient approximately two orders of magnitude greater than absorption plays a dominant role. The potential of an experimental laser diode based setup for the transillumination of rheumatoid finger joints and the pattern of the stray light detection are demonstrated. For evaluating the scattering light images a new non-local image segmentation method is presented. Regarding a noisy picture as a multicomponent mixture of gray scaled particles, this method minimizes a non-convex free energy functional under the constraint of mass conservation of the components. Contrary to constructing equilibrium distributions as steady states of an adequate evolution equation, a direct descent method for the free energy is used to separate the components of the image.