Fig 1.
Illustration of the main principle behind the proposed US texture characterization approach.
Fig 2.
Main steps of the general concept of the signal processing algorithm for texture modelling and feature extraction in US images.
Fig 3.
Conversion of a matrix by traversing the matrix and its transposed in ZigZag and in spiral.
Fig 4.
Example of a CWT decomposition of a thyroid US image when three image patches are taken from different locations of the US image.
Fig 5.
AR spectra for the patches IUS1, IUS2 and IUS3 (in blue, red and black lines respectively) for the four narrowband signals belonging to the ZigZag matrix to signal conversion.
Fig 6.
Color-map of the computed features in patches belonging to thyroid and three classes of non-thyroid regions.
Table 1.
Spectra used in the numerator (NUM) and denominator (DEN) of Eq (3) for computing the 30 energy ratio features.
Fig 7.
Mean and standard deviation of values of ERs features 1 to 15 of thyroid and non-thyroid patches for the 6 subjects of the Dataset 1.
Fig 8.
Mean and standard deviation of values of ERs features 16 to 30 of thyroid and non-thyroid patches for the 6 subjects of the Dataset 1.
Fig 9.
Example of obtained AR spectral energy ratios when the approach is applied to the full set of patches extracted from the thyroid US Dataset 1.
Fig 10.
Examples of thyroid segmentation using the proposed approach and comparison with the ground truth.
Table 2.
Comparison of the proposed approach in terms of Dice Coefficient using the Dataset 1 with algorithms compared in [8].
Table 3.
Comparison of the proposed approach using Dataset 2 with five algorithm results reported in [9].