Real-time bilateral filtering of ultrasound images through highly optimized DSP implementation

The examination of anatomical structures through ultrasound (US) is widely employed in everyday medical diagnostics. Unfortunately, the quality achievable in US imaging is limited by speckle noise. This sort of noise produces a particular spotted pattern which makes image interpretation difficult, hampers the application of automatic segmentation methods, and can hide small but important morphological structures. Several methods have been developed for reducing the effect of speckle noise without interfering with image information. Bilateral filtering has been proven to produce good performance, but unfortunately, it requires a high calculation effort which hampers its real-time employment. In this work we present a real-time implementation of Bilateral filtering for the TMS320C64x+ Digital Signal Processors. The real-time capability has been achieved by careful code optimization and efficient exploitation of the DSP architecture. The filter, tested on the ULA-OP research scanner, is shown suitable to process 192x512 images at 40 fps. Images obtained on both phantom and in-vivo are presented.