Spatial and temporal speckle filtering for feature extraction in ultrasound images
Ultrasound is a low cost, non-invasive imaging modality that has proved popular for many medical applications. However, the coherent nature of ultrasound results in images that are corrupted by speckle noise which reduces the effectiveness of ultrasound as a diagnostic tool. This thesis concentrates on extending the filtering techniques for reducing the speckle content of ultrasound images to produce images which are more suited to automatic interpretation (a topic of increasing interest) with the advantage of retaining the original, unfiltered, image. The Truncated Median Filter can be used to filter a single image by replacing each point with an output which is equivalent to the mode of the local distribution. This corresponds to a maximum likelihood estimate for a signal contaminated by speckle in a Rayleigh distribution. Unlike extant speckle filtering methods the Truncated Median Filter can be implemented fully automatically. This thesis demonstrates that not only does the Truncated Median Filter have theoretical advantage over other techniques but also that its results demonstrate edge preservation, together with speckle reduction, which have advantages for automatic feature extraction. A fresh approach has been adopted to analyse the filtering results that objectively assesses the suitability of images for a further image processing stage. This contrasts with previous evaluation schemes that have predominantly been concerned with the diagnostic attributes of images. Objective evaluation is achieved by closely considering the result of filtering on edges, and hence on feature boundaries, which are of critical importance to later feature extraction techniques. The evaluation scheme uses simulated, test object and in vivo ultrasound images to assess the filtering results. Simulated images were used since the exact speckle characteristics are known and can be controlled.