Real-Time US Image Enhancement by Forward-Backward Diffusion Using GPU

The forward-backward diffusion (FBD) can be used for edge improvement in any image. The case considered in this paper is that of ultrasound (US) images. The US images are imperfect since the useful signal representing anatomical details is obscured by speckles inherent in this approach. A method such as the nonlinear coherent diffusion (NCD) can remove speckles to a large extent. However, the NCD-processed image can be further enhanced by the FBD. This paper describes the details of the proposed GPU implementation of the FBD algorithm. The obtained execution time of one iteration of the FBD algorithm is less than 2.2 ms for 1 MB image.