Primary Rectal Cancer: Repeatability of Global and Local-Regional MR Imaging Texture Features

Purpose To assess the day-to-day repeatability of global and local-regional magnetic resonance (MR) imaging texture features derived from primary rectal cancer. Materials and Methods After ethical approval and patient informed consent were obtained, two pretreatment T2-weighted axial MR imaging studies performed prospectively with the same imaging unit on 2 consecutive days in 14 patients with rectal cancer (11 men [mean age, 61.7 years], three women [mean age, 70.0 years]) were analyzed to extract (a) global first-order statistical histogram and model-based fractal features reflecting the whole-tumor voxel intensity histogram distribution and repeating patterns, respectively, without spatial information and (b) local-regional second-order and high-order statistical texture features reflecting the intensity and spatial interrelationships between adjacent in-plane or multiplanar voxels or regions, respectively. Repeatability was assessed for 46 texture features, and mean difference, 95% limits of agreement, within-subject coefficient of variation (wCV), and repeatability coefficient (r) were recorded. Results Repeatability was better for global parameters than for most local-regional parameters. In particular, histogram mean, median, and entropy, fractal dimension mean and standard deviation, and second-order entropy, homogeneity, difference entropy, and inverse difference moment demonstrated good repeatability, with narrow limits of agreement and wCVs of 10% or lower. Repeatability was poorest for the following high-order gray-level run-length (GLRL) gray-level zone size matrix (GLZSM) and neighborhood gray-tone difference matrix (NGTDM) parameters: GLRL intensity variability, GLZSM short-zone emphasis, GLZSM intensity nonuniformity, GLZSM intensity variability, GLZSM size zone variability, and NGTDM complexity, demonstrating wider agreement limits and wCVs of 50% or greater. Conclusion MR imaging repeatability is better for global texture parameters than for local-regional texture parameters, indicating that global texture parameters should be sufficiently robust for clinical practice. Online supplemental material is available for this article.


Kurtosis
Describes the "peak" of a distribution. Kurtosis > 3: sharper peak than a normal distribution. Kurtosis < 3: flatter peak than a normal distribution. Kurtosis = 3: normal distribution. Where is a vector that contains the histogram counts and is total number of different gray levels present in the image. Entropy Measures voxel randomness. Low entropy is noted in homogeneous voxels.
Where is a vector that contains the histogram counts and is total number of different gray levels present in the image.
Note.-These statistics provide an indication of central tendency (mean, median) and variability (skewness, kurtosis, energy, and entropy). Where i is the voxel value betw een i = 1 and imax in the region of interest, j is the voxel value betw een j = 1 and jmax in the region of interest, and p(i,j) is the probability of the occurrence of that voxel value i relative to j.

Homogeneity
Measures the relative homogeneity betw een voxels. The value is high for homogeneous images.
Where p(i,j) is the probability of the occurrence of that voxel value i relative to j.

EnergyGLCM
Measures homogeneity. Homogeneous images demonstrate high energy. p(i,j) is the probability of the occurrence of that voxel value i relative to j.

ContrastGLCM
Measures local gray-level intensity variation and favors contributions from voxels p(i, j) away from the diagonal of (ie, i ≠ j).

 
Where i is the voxel value betw een i = 1 and imax in the region of interest, j is the voxel value betw een j = 1 and jmax in the region of interest, n is the number of voxels, and p(i,j) is the probability of the occurrence of that voxel value i relative to j.
is the number of distinct gray levels in the quantized image. Autocorrelation Measures gray-level intensity linear dependence betw een the voxels (i,j) at the specified positions relative to each other.

  ( , )
ij ij p i j  , Where i is the voxel value betw een i = 1 and imax in the region of interest, j is the voxel value betw een j = 1 and jmax in the region of interest, and p(i,j) is the probability of the occurrence of that voxel value i relative to j.

Cluster shade
Measures the skew ness of the matrix and is thought to gauge the perceptual concepts of uniformity. When the cluster shade is high, the image is more skew ed.
Where i is the voxel value betw een i = 1 and imax in the region of interest, j is the voxel value betw een j = 1 and jmax in the region of interest, and p(i,j) is the probability of the occurrence of that voxel value i relative to j. µ x Is the mean of x , w hereas µ y is the mean of y . x Is obtained by summing row s of . that is, ∑ ( , ), and y is obtained by summing columns of . that is, . is the number of distinct gray levels in the quantized image.
Cluster prominence Measures asymmetry. A low cluster prominence value indicates small variations in grayscale.
Where i is the voxel value betw een i = 1 and imax in the region of interest, j is the voxel value betw een j = 1 and jmax in the region of interest, p(i,j) is the probability of the occurrence of that voxel value i relative to j, µx is the mean of px, and µy is the mean of py. x Is obtained by Page 3 of 9 summing row s of . that is, ∑ ( , ) and y is obtained by summing columns of . that is,

Difference entropy
Measure of the inhomogeneity of the image. High difference entropy reflects greater inhomogeneity.
Ng is the number of quantized gray levels in image G, i is the voxel value betw een i = 1 and imax in the region of interest, and p(x-y) is the probability of the occurrence of that voxel value x relative to y.

Difference variance
Measure of the inhomogeneity of the image. High difference variance reflects greater inhomogeneity.
Where Ng is the number of quantized gray levels in image G, i is the voxel value betw een i = 1 and imax in the region of interest, p(x-y) is the probability of the occurrence of that voxel value x relative to y, and µx-y is the mean of px-y.

Dissimilarity
Measure of the inhomogeneity of the image. Dissimilarity is a high value for inhomogeneous images and a relatively low er value for homogeneous images.
Where i is the voxel value betw een i = 1 and imax in the region of interest, j is the voxel value betw een j = 1 and jmax in the region of interest, and p(i,j) is the probability of the occurrence of that voxel value i relative to j. Inverse difference moment Measures the relative homogeneity betw een voxels.
Where i is the voxel value betw een i = 1 and imax in the region of interest, j is the voxel value betw een j = 1 and jmax in the region of interest, and p(i,j) is the probability of the occurrence of that voxel value i relative to j.

Maximum probability
Measures the maximum probability of pixel combination. High values occur if one combination of pixels dominates the pixel pairs in the image.

max ( , )
Where p(i,j) is the probability of the occurrence of that voxel value i relative to j.  Where Ng is the number of quantized gray levels in image G and x +y ( ) = ∑ ∑ (x, y) y x where x + y = and = 2,3, … . , 2

Sum variance
Measure of the inhomogeneity of the image. High sum variance reflects greater inhomogeneity.
Where Ng is the number of quantized gray levels in image G, i is the voxel value betw een i = 1 and imax in the region of interest.
x +y ( ) = ∑ ∑ (x, y) y x where x + y = and = 2,3, … . , 2 Note.-An original texture image, D, is requantized into an image G with a reduced number of gray levels, Ng. A typical value of Ng is 16 or 32. Then, GLCM is computed from G by scanning the intensity of each voxel and its neighbor, defined by displacement d and angle . Displacement d could take a value of 1,2,3,…n, whereas angle  was limited to 0, 45, 90 and 135. The GLCM   , | ,θ p i j d is a second-order joint probability density function of gray-level pairs in the image for each element in the co-occurrence matrix obtained by dividing each element with Ng. Finally, scalar secondary features are extracted from this co-occurrence matrix.  , Where N is the number of gray levels, gk is the gray level w ithin the voxel, pg is the histogram of the gray-level difference at specific distance d, and d is the displacement vector. Variance Measures the dispersion of gray-level differences at a certain distance, d.
Where µd is the mean difference, N is the number of gray levels, gk is the gray level w ithin the voxel, pg is the histogram of the gray-level difference at specific distance d, and d is the displacement vector.

ContrastGLDM
Measures the local variations in gray-level difference in the image.
Where N is the number of gray levels, gk is the gray level w ithin the voxel, pg is the histogram of the gray-level difference at specific distance d, and d is the displacement vector.
Note.-GLDM = gray-level difference matrix. The gray-level differences are computed by taking the absolute differences of all possible pairs of gray levels distance d apart at angle  and counting the number of times the difference is 0,1, …,255. d Is (dx,dy), the displacement vector between two image pixels, and g(d) is the gray-level difference at distance d. pg (g,d) Is the histogram of the gray-level differences at the specific distance d. One distinct histogram exists for each distance d. The difference statistics are then normalized by dividing each element of the vector by the number of possible pixel pairs. Se veral texture measures can be extracted from the histogram of gray-level differences. N is the number of gray levels.

Parameter
Description Formula

Coarseness
Measures gray-tone difference within the image. High coarseness represents areas where gray-tone differences are small. Small gray-tone differences possess a high degree of local uniformity in intensity.
Where s is the NGTDM vector, pi is the probability of a voxel value for voxels, n is the number of unique gray levels present in an image, and Gh is the highest gray level in an image.

ContrastNGTM
Measures local variation in intensity. High contrast indicates areas of large intensity difference between neighboring regions.
Where s is the NGTDM vector; pi is the probability of a voxel value for voxels, n is the number of unique gray levels present in an image, and Gh is the highest gray level in an image.

Busyness
Measures changes in intensity from one voxel to its neighbor; high busyness indicates that the spatial frequency of intensity changes is very high. , Where s is the NGTDM vector, pi is the probability of a voxel value for voxels, n is the number of unique gray levels present in an image, and Gh is the highest gray level in an image.
Note.-For a given three-dimensional image, a GLSZM Z is defined as follows: Each element z(i,j) represents the number of zones with pixels of gray-level intensity equal to i and size of zone equal to j. The size of the matrix Z is M  N, where M is the maximu m gray level in the scan and N is equal to the possible maximu m zones in the corresponding s can.  Measures the standard deviation of a fractal computed by a differential box-counting algorithm. Measures the amount of "gaps" in the image or object. If a fractal has large gaps, it has high lacunarity. Measures the density of the image or object (ie, how much the image or object occupies the space that contains it). A small value corresponds to a coarse texture.