Structure visualization and identification for ultrasound (US) arthral structure images is essential in US-guided arthrosis operation and diagnosis for arthrosis disorder. US imaging can image arthrosis tissues and cartilaginous, but it often depicts the structure poorly, because of speckle, reverberation, shadowing feature and other artifacts. The enhancement of arthrosis structure visualization remains great challenge in US images. This paper proposes an improved diffusion to remove reverberation noise, smoothing speckles and improve the visualization of arthrosis structure in US images. The structure tensor from automatic detection is applied to keep edges structure. The potential of the method is demonstrated with the results compared to nonlinear diffusion. The preliminary results show that improved diffusion is a robust method to enhance surfaces visualization of US arthrosis structure. The method shows promise in a clinical application of arthrosis operation planning and arthrosis diagnosis.