Prostate Tissue Characterization via Ultrasound Speckle Statistics

In this work we study methodologies for speckle extraction and analysis in ultrasound biomedical images. Assuming a multiplicative noise model, the investigated methods exploit the decorrelating properties of the wavelet transform for non-stationary signals. The efficiency of preprocessing procedures which decompose the acquired signal into coherent and diffuse component is investigated. The different approaches are evaluated in terms of computational cost and effectiveness in tissue characterization of human prostates affected by carcinoma. In particular, we compare the performances of fractal and statistical features for the classification of textures. By analyzing speckle statistics we obtain a fundamental tissue "signature" suitable for image segmentation and characterization