Unsupervised segmentation of RF echo into regions with differentscattering characteristics

摘要

Recent experimental results verify that the probabilityndistribution function of the diffuse component of the RF echo dependsnprimarily on the concentration of the diffuse scatterers in thenresolution cell. In this paper we apply these results to develop annunsupervised segmentation scheme that partitions an RF A-scan or B-scannimage into statistically homogeneous regions that reflect the underlyingnscattering characteristics. The proposed segmentation scheme is based onna nonparametric homogeneity test that compares two regions of interestn(ROI) for possible merging utilizing information about both the coherentnand the diffuse component of the RF echo. For the coherent component,nhomogeneity is defined in terms of the estimated average spacing of eachnROI. For the diffuse component, we use the nonparametricnKolmogorov-Smirnov (K-S) homogeneity statistical test that compares twonempirical distributions associated with any two ROIs. This test can benused to obtain a segmentation into regions with different scatteringncharacteristics regardless of the nature of the scattering conditionsn(e.g., Rayleigh regions with different scatterer concentration,ndifferent non-Rayleigh regions, or different coherent scatteringnregions). Finer segmentation can be obtained by learning thendistributions associated with the various homogeneous regions obtainednfrom the coarse segmenter. The proposed segmentation scheme is appliednon simulated RF scans with different scatterer concentration pernresolution cell, on phantom data which mimic tissue, and on liver scans.nThe results demonstrate the effectiveness of the segmentation algorithmneven in cases of subtle differences in the scattering characteristics ofneach region (for example, diffuse component with scatterer density of 16nand 32 scatterers per resolution cell)