Description of Texture Features Employed in This Work

Supplementary Mmaterial

Description of texture features employed in this work:

Haralick texture features: Haralick features are based on quantifying the spatial gray-level co-occurrence within local neighborhoods around each pixel in an image, stored in the form of matrices. A total of 13 Haralick texture descriptors were calculated from every lesion for every sequence by computing median of the statistics derived from the corresponding co-occurrence matrices.

Laws Texture features: Laws features use 5x5 separable masks that are symmetric or anti-symmetric to extract level (L), edge (E), spot (S), wave (W), and ripple (R) patterns in an image. The convolution of these masks with every image resulted in a total of 25 distinct Laws features for each image for every individual MRI sequence.

Laplacian pyramids: Laplacian pyramids allow for capturing multi-scale edge representations via a set of band pass filters. First, the original image is convolved with a Gaussian kernel. The Laplacian is then computed as the difference between the original image and the low pass filtered image. The resulting image is then sub-sampled by a factor of two, and the filter sub-sample operation is repeated recursively. This process is continued to obtain a set of band-pass filtered images (since each is the difference between two levels of the Gaussian pyramid). A total of 24 filtered image representations are obtained from every lesion for every MRI sequence by computing the median of feature values across all pixels within a lesion.

Histogram of gradient orientations: For every pixel c on an image, gradients along X and Y direction are computed as, and. The gradient orientation is then computed as, . After obtaining the gradient orientation at every pixel, c, within the segmented lesion, they are binned into a histogram, spanning from to. The entire histogram was binned in 20 bins, each bin spanning 18°. The feature vector consists of binned histogram values in the form of vectors of length 20 X 1.

Effect of intensity standardization on Gd-T1 MRI scans:

Figure 1I: Illustration of intensity drift across multi-institutional data for T1wMRI, by plotting the distributions of randomly chosen 10 patient studies along the same axis before (a) and after intensity standardization (b). Note that after intensity standardization, the distributions across studies from different institutions are no longer misaligned, suggesting successful correction of the drift artifact.