PET Radiomics in NSCLC: state of the art and a proposal for harmonization of methodology

Imaging with positron emission tomography (PET)/computed tomography (CT) is crucial in the management of cancer because of its value in tumor staging, response assessment, restaging, prognosis and treatment responsiveness prediction. In the last years, interest has grown in texture analysis which provides an “in-vivo” lesion characterization, and predictive information in several malignances including NSCLC; however several drawbacks and limitations affect these studies, especially because of lack of standardization in features calculation, definitions and methodology reporting. The present paper provides a comprehensive review of literature describing the state-of-the-art of FDG-PET/CT texture analysis in NSCLC, suggesting a proposal for harmonization of methodology.


Statistical-based 31
Shape and size 1,2 33 34 To describe the shape and size of the volume of interest (V = the volume and A, the surface area of the 35

volume of interest.) 36
Asphericity: is a quantitative measure of shape irregularity caused by necrotic tumor parts or invasive 37 growth.  The gray-level co-occurrence matrix features (GLCM) (gray level = gray tone) 94 The i,j th element of the co-occurrence matrix for an anatomical structure of interest represents the number of 95 times that intensity levels i and j occur in two voxels separated by distance (d) in direction (a). The co-96 occurrence features are based on the second-order joint conditional probability density function P(i,j; a,d) of 97 a given texture image. 98 These metrics are independent of tumor position, orientation, size, and intensity and take into account the 99 spatial distribution of the local voxel intensities.

Gray-level size-zone matrix-based features (GLSZM) 145
Let: 146 p(i, j) be the (i, j) th entry in the given size-zone matrix p, 147 Ng the number of discrete intensity values in the image, 148 Nz the size of the largest, homogeneous region in the volume of interest, 149 Small-area emphasis (SAE) or short-zone emphasis (SZE): Large-area emphasis (LAE) or long-zone emphasis (LZE): Intensity variability (IV): Gray-level non-uniformity for zone (GLUNz): Size-zone variability (SZV): Busyness: For the gradient feature calculation the following neighborhood for image pixel x(i,j) is defined: 199 where h is the lag vector, N is the number of cases, x i (target or head voxel value) is the variable 234 value at a particular location i, x j (source or tail voxel value) is the variable value at another 235 location, X is the mean of the variable, and w ij is a weight applied to the comparison between location I and location j. In more current use, w ij is a distance-based weight which is the inverse distance between locations i and j.
deviations from the mean but the deviation in intensities of each observation location with another 240 one. Its formal definition is: The values of C typically vary between 0 and 2. The theoretical value of C is 1, which indicates that The linear dependence that one pixel of an image has on another is well known and can be illustrated by the 253 autocorrelation function. 254 A pixel (i,j ) depends on a two-dimensional neighborhood N(i,j) consisting of pixels above or to the left of it 255 as opposed to the simple sequence of the previous pixels a raster scan could define. For each pixel (k,I ) in an 256 order-D neighborhood for pixel (i, j), (k,I ) must be previous to pixel (i,j ) in a standard raster sequence and 257 (k , I) must not have any coordinates more than D units away from(i,j). Formally, the order-D neighborhood 258 where k is a scaling constant and D is the fractal dimension that is used to detect self-affinity. The Fractal dimension and fractal abundance were calculated using a box-counting method, with multiple grid 295 offsets for all possible box start locations, based on the following equation: 296 where L is the box size, N L is the number of boxes of size L needed to cover the object being studied, and D 298 is the fractal dimension. By plotting a log-log plot of N L versus L, fractal dimension (FD) can be obtained 299 from the slope, and fractal abundance (FA) or log K can be obtained from the y-intercept of the straight 300 portion of the curve. where Λ represents lacunarity, r represents box size, s represents the number of occupied sites within a box 306 size, and Q is the probability distribution (representing the frequency distribution of the total number of 307 occupied sites for a box size r over the total number of boxes of size r). Lacunarity is displayed by plotting a 308 log-log plot of lacunarity versus size of the gliding box. Lacunarity thus provides a dimensionless 309 representation of the fraction of sites that is occupied.

348
The general mother wavelet can be constructed from the following scaling φ(x) and wavelet