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.

Prognostic value of pre-therapeutic FDG-PET radiomic analysis in gastro-esophageal junction cancer

“…This confirms our findings about Histogram-Entropy that was highly correlated to GLCM-Entropy, appearing to be a robust TF to assess prognostic value of esophageal cancers. However, to avoid the lack of reproducibility and to allow harmonization of practice, recent standardization of radiomic procedures has been proposed 30 , 31 . The future of radiomics will undeniably involve artificial intelligence, which will further strengthen its robustness 32 .…”

“…Secondly, our study was retrospective and conducted in a single center with a risk of selection bias. However, studying PET radiomic in multicentric trials remains difficult in terms of methodology, even if harmonization procedures between the different machines are available to limit the variability resulting from different image resolutions and reconstructions 31 , 33 . Thirdly, there is near 10% of patients lost to follow-up and follow-up is only 3 years; the disease has a poor prognosis so there are few survivors beyond.…”

A Critical Analysis of the Robustness of Radiomics to Variations in Segmentation Methods in 18F-PSMA-1007 PET Images of Patients Affected by Prostate Cancer

“…Moreover, the analysis showed that wavelet decomposition is the worst performing filter in terms of reproducibility, with 47.85% wavelet features exhibiting poor reproducibility. It has been shown in previous studies [63,64] that image denoising could lead to more robust features, and the difference in performance between LoG and wavelet could be due to the different wavelet sub-band combinations. Indeed, the combinations that started with a high pass filter (HHH, HLH, HHL, HLL) obtained the lowest average ICC (ICC HHL : 0.34), while those that started with a low pass filter obtained the highest values, with the LLL sub-band showing the highest average ICC (0.8).…”

AAPM task group report 273: Recommendations on best practices for AI and machine learning for computer‐aided diagnosis in medical imaging

“…Image‐domain harmonization methods include post‐processing of image data 45 and style transfer, 46 and feature‐domain harmonization methods include basic statistical normalization techniques 47 and advanced statistical techniques such as ComBat 48,49 . The Quantitative Imaging Biomarkers Alliance (QIBA) and the Quantitative Imaging Network (QIN) have also devoted efforts to the harmonization of medical imaging data and tools 50,51 . It is important to recognize that although data harmonization aims to reduce the systematic variations due to image acquisition, reconstruction, and post‐processing or due to different protocols among data collection sites, it does not address the issue of systematic variations among patient sub‐populations (see sections 2.1.3.2 and 4.2.2.3).…”