FILTERING AND RESTORATION OF STRUCTURES IN 3D ULTRASOUND IMAGES

We present a new method aimed at restoring structures in 3D US images. In our approach, 3D US data acquired with tilt devices is resampled into cylindrical coordinates with the purpose of overcoming the problems of anisotropic sampling space. Then, enhancement and shadowing effects, as well as average attenuation effects, are removed with a ray-based rescaling process. Finally, a novel method for reducing noise and enhancing the structures based on a modified 3D anisotropic diffusion is applied. The diffusion scheme is improved in several steps: the diffusivity is computed locally as part of each iteration based on local statistics; two terms intended to preserve the mean value along homogeneous regions and to enhance the contrast are introduced. Results show an improvement of the contrast when applied on 3D transrectal US (TRUS) images of prostate, which can facilitate further segmentation.