Optimization of Ultrasound Image Despeckling Using Speckle Size Estimation

The presence of speckle noise poses challenges in the interpretation of diagnostic medical ultrasound (US) images, particularly in detecting low-contrast structures. Among the traditional despeckling approaches, the Lee filter aims at removing this noise while preserving the useful edges of the image. We propose a method to optimize the design process of the Lee filter for the removal of speckle noise from diagnostic medical ultrasound images. Our hypothesis is that it is possible to improve the performance of the Lee filter by adapting the size of its window to the characteristic size of the speckle noise. To test this possibility, we explored a possible relationship between speckle size and the optimal filter window size. We evaluated a measure of speckle dimension based on the estimation of the image auto-covariance function. Then two image quality indexes (Peak Signal-to-Noise Ratio and Structural Similarity Index) were used to assess filter efficiency at different window sizes. Lastly, we implemented a multinomial logistic regression model to generalize this association. This work sets the stage for the possible application of speckle size assessment in optimizing the design of spatial filters. Furthermore, through the analysis of the speckle size in filtered images, we propose to exploit this characteristic to help define an unsupervised measure of image Quality.