Identification of robust and reproducible CT‐texture metrics using a customized 3D‐printed texture phantom

Abstract Objective The objective of this study was to evaluate the robustness and reproducibility of computed tomography‐based texture analysis (CTTA) metrics extracted from CT images of a customized texture phantom built for assessing the association of texture metrics to three‐dimensional (3D) printed progressively increasing textural heterogeneity. Materials and Methods A custom‐built 3D‐printed texture phantom comprising of six texture patterns was used to evaluate the robustness and reproducibility of a radiomics panel under a variety of routine abdominal imaging protocols. The phantom was scanned on four CT scanners (Philips, Canon, GE, and Siemens) to assess reproducibility. The robustness assessment was conducted by imaging the texture phantom across different CT imaging parameters such as slice thickness, field of view (FOV), tube voltage, and tube current for each scanner. The texture panel comprised of 387 features belonging to 15 subgroups of texture extraction methods (e.g., Gray‐level Co‐occurrence Matrix: GLCM). Twelve unique image settings were tested on all the four scanners (e.g., FOV125). Interclass correlation two‐way mixed with absolute agreement (ICC3) was used to assess the robustness and reproducibility of radiomic features. Linear regression was used to test the association between change in radiomic features and increased texture heterogeneity. Results were summarized in heat maps. Results A total of 5612 (23.2%) of 24 090 features showed excellent robustness and reproducibility (ICC ≥ 0.9). Intensity, GLCM 3D, and gray‐level run length matrix (GLRLM) 3D features showed best performance. Among imaging variables, changes in slice thickness affected all metrics more intensely compared to other imaging variables in reducing the ICC3. From the analysis of linear trend effect of the CTTA metrics, the top three metrics with high linear correlations across all scanners and scanning settings were from the GLRLM 2D/3D and discrete cosine transform (DCT) texture family. Conclusion The choice of scanner and imaging protocols affect texture metrics. Furthermore, not all CTTA metrics have a linear association with linearly varying texture patterns.


h. Skewness (SKEW):
Skewness is a measure of the lack of symmetry in a distribution. A symmetric distribution has a skew of zero. A positive skewness indicates a positively skewed distribution and likewise a negative skewness indicates a negatively skewed distribution.  Histogram analysis is completely based on the distribution of the grayscale values forming the ROI; it provides no information about the spatial relationship of the pixels to each other. Therefore, differentiating 2 completely different texture patterns with the same number but different orientation of black and white pixels is not possible in histogram analysis.
In addition, 4 size-based metrics are calculated: number of voxel dimensions in x, y, and z directions and the volume.
Also called sum of squares variance, CON weights pixels in the GLCM/GLDM map exponentially more as their distance from the diagonal increases. A larger value indicates greater variations in gray levels compared to their neighborhood.

d.
Dissimilarity (DIS): DIS is similar to contrast, except that the weighting scheme is linear compared to exponential.

e.
Homogeneity (HOM): HOM weights pixels in the GLCM/GLDM map exponentially less as their distance from the diagonal increases. A larger value indicates smaller variations in gray levels compared to their neighborhood. This is an inverse metric of CON.

f.
Inverse Difference Moment (IDM): IDM is similar to homogeneity, except that the weighting scheme is linear compared to exponential.
Inverse Difference Moment Normalized (IDMN): IDMN is similar to IDM, except that the weighting scheme is exponential, and it is a measure of the local homogeneity of an image. l.

Sum of average (SUMAVG):
For an image of single color of no variation, the sum of average values for different angles are 2. Usually, for an image of varied pixel values, the sum of average is high in value.
Sum of entropy (SUMENT): As per the definition of entropy, this value goes higher for an image of more variations.
Sum of variance (SUMVAR): Usually for an image of varied pixel values, the sum of variance is high in value.
Difference of average (DIFAVG) and difference of entropy (DIFENT) is calculated similarly to sum of average and entropy respectively, except, differences are calculated instead of summations.

p.
Standard deviation (SD) and mean are same as those calculated for histogram analysis, except run on GLCM/GLDM maps.

s.
Root mean square (RMS): RMS computes the root mean square value of each row or column of the input, along vectors of a specified dimension of the input, or of the entire input.

Fast Fourier Transform (FFT) Analysis (3 metrics):
Specifically, a 512-point FFT was applied to all tumor images. Using the built-in MATLAB implementation of the FFT algorithm (FFT2), we extracted the individual frequencies, amplitude (how much frequency of a given type is present), and phase (where in the image the frequency is present) of the original image. The resultant magnitude and phase of the FFT across all images were analyzed. Three metrics were defined. In all cases, the harmonics analysis was limited between 15% and 95% of maximum spatial frequency within the tumor. These cutoffs were chosen to avoid inclusion of low-frequency content (i.e. tumor size effect and noise) and high-frequency noise. The selected band-pass frequencies correspond to the spatial frequencies within the tumor.
a. Entropy of FFT magnitude (E_FFT_Mag): Diversity (Randomness) measure in the magnitude of FFT harmonics.
where Pk is each harmonic from the FFT transformed (magnitude) tumor image.
The E_FFT_Mag of a homogenous texture should be smaller compared to that of a heterogeneous texture.

b.
Entropy of FFT phase (E_FFT_Phase): Diversity (Randomness) measure in the phase of FFT harmonics.
where Pk is every harmonic from the FFT transformed (phase) tumor image.
The E_FFT_Phase of a homogenous texture should be smaller compared to that of a heterogeneous texture. where Pk is every harmonic from the FFT transformed (amplitude) tumor image.
The CI of a homogenous texture should be a smaller value compared to the CI of a heterogeneous texture.   Gray Level Variance measures the variance in zone counts for the gray levels.

Two-dimensional and Three-dimensional Gray
Size Zone Variance measures the variance in zone counts across different zone sizes.

p.
Ng: Total number of gray levels in the image.

q.
Ns: Total number of zone sizes in the image.

r.
Np: Total number of voxels in the image.

Neighboring Gray Tone Difference Matrix (NGTDM) (6 metrics):
An NGTDM quantifies the difference between a gray value and the average gray value of its neighbors within distance δ.