Noise-Based Image Harmonization Significantly Increases Repeatability and Reproducibility of Radiomics Features in PET Images: A Phantom Study

Abstract

For multicenter clinical studies, characterizing the robustness of image-derived radiomics features is essential. Features calculated on PET images have been shown to be very sensitive to image noise. The purpose of this work was to investigate the efficacy of a relatively simple harmonization strategy on feature robustness and agreement. A purpose-built texture pattern phantom was scanned on 10 different PET scanners in 7 institutions with various different image acquisition and reconstruction protocols. An image harmonization technique based on equalizing a contrast-to-noise ratio was employed to generate a "harmonized" alongside a "standard" dataset for a reproducibility study. In addition, a repeatability study was performed with images from a single PET scanner of variable image noise, varying the binning time of the reconstruction. Feature agreement was measured using the intraclass correlation coefficient (ICC). In the repeatability study, 81/93 features had a lower ICC on the images with the highest image noise as compared to the images with the lowest image noise. Using the harmonized dataset significantly improved the feature agreement for five of the six investigated feature classes over the standard dataset. For three feature classes, high feature agreement corresponded with higher sensitivity to the different patterns, suggesting a way to select suitable features for predictive models.

Keywords: PET radiomics features; feature agreement; image harmonization; repeatability; reproducibility.

Figure 1

Texture phantom used for this study. (A) Photographs of the manufactured cylindrical texture compartments inserted into a standard NEMA phantom shell. (B) Each compartment mimics a different distribution of radiotracer uptake due to a different structural pattern of blocked space. (C) Axial and coronal CT and PET images of the phantom.

Figure 2

Single slices of the standard (A) and harmonized (B) datasets. The order of the images follows the order in Table 1 (scanner 1 to 10 from top left to bottom right). The images were normalized by the mean activity of their respective background (Pattern-ROI-5). The displayed color scale is identical for all images and ranges from 0 (black) to 1.22 (red).

Figure 3

Intraclass coefficient ICC(1,1) for all radiomics features from the PET images in the repeatability dataset. For each feature, three values of ICC(1,1) are plotted: binning time 10 min (green), 5 min (red), and 2 min (black). Error bars indicate the 95% confidence intervals of the ICC values. The results are sorted in descending value of ICC(1,1) for the 2 min binning time (black circles). The feature class of each feature is indicated by the color of the tiles (dark blue: first order, light blue: GLCM, cyan: GLDM, yellow: GLRLM, orange: GLSZM, brown: NGTDM).

Figure 4

ICC(2,1) values for all features for the reproducibility datasets S10, H10, S6, and H6 (Table 4), ranked by decreasing value of the H10 dataset (solid black). The feature class of each feature is indicated by the color of the tiles (dark blue: first order, light blue: GLCM, cyan: GLDM, yellow: GLRLM, orange: GLSZM, brown: NGTDM).

Figure 5

ICC(2,1) values grouped by radiomics feature class. A star (*) indicates statistical significance according to the paired Wilcoxon signed-rank test.

Figure 6

Feature reproducibility across different scanners (ICC(2,1)) versus pattern sensitivity (inter-ROI standard deviation) for all features in the standard (A) and harmonized (B) datasets.