Identifying Robust Radiomics Features for Lung Cancer by Using In-Vivo and Phantom Lung Lesions

Abstract

We propose a novel framework for determining radiomics feature robustness by considering the effects of both biological and noise signals. This framework is preliminarily tested in a study predicting the epidermal growth factor receptor (EGFR) mutation status in non-small cell lung cancer (NSCLC) patients. Pairs of CT images (baseline, 3-week post therapy) of 46 NSCLC patients with known EGFR mutation status were collected and a FDA-customized anthropomorphic thoracic phantom was scanned on two vendors’ scanners at four different tube currents. Delta radiomics features were extracted from the NSCLC patient CTs and reproducible, non-redundant, and informative features were identified. The feature value differences between EGFR mutant and EGFR wildtype patients were quantitatively measured as the biological signal. Similarly, radiomics features were extracted from the phantom CTs. A pairwise comparison between settings resulted in a feature value difference that was quantitatively measured as the noise signal. Biological signals were compared to noise signals at each setting to determine if the distributions were significantly different by two-sample t-test, and thus robust. Four optimal features were selected to predict EGFR mutation status, Tumor-Mass, Sigmoid-Offset-Mean, Gabor-Energy and DWT-Energy, which quantified tumor mass, tumor-parenchyma density transition at boundary, line-like pattern inside tumor and intratumoral heterogeneity, respectively. The first three variables showed robustness across the majority of studied CT acquisition parameters. The textual feature DWT-Energy was less robust. The proposed framework was able to determine robustness of radiomics features at specific settings by comparing biological signal to noise signal. Identification of robust radiomics features may improve the generalizability of radiomics models in future studies.

Robustness of pulmonary nodule radiomic features on computed tomography as a function of varying radiation dose levels—a multi-dose in vivo patient study

“…We found that firstorder features were neither more robust nor more correctable by a linear model than other features. This is in contradiction to Hepp et al and Kim et al who found that first-order features were among the most stable in, respectively, a noise simulation study and a phantom study [10,17].…”

“…Although some phantom studies have shown that the effect of varying tube current on radiomic features does not significantly affect radiomic features [ 6 ], other studies have shown milliampere-second variation does in fact significantly influence radiomic feature values [ 7 , 8 ]. Although several in vivo dose modulation radiomic feature robustness studies have been performed to date, these studies are retrospective in the sense that they compare features taken from a single diagnostic scan, and later follow-up scans [ 9 , 10 ]. As mentioned in the systematic review by Reiazi et al: “The drawbacks of the retrospective studies are that the investigators did not have control over the parameters studied, and the range of the scan acquisition parameter variations were limited to those used in imaging patients.” [ 11 ].…”

Artificial intelligence-driven radiomics study in cancer: the role of feature engineering and modeling

“…Several studies have been performed to select features with high repeatability and reproducibility for use in feature engineering. Lu et al [ 99 ] developed a new phantom-based framework to screen radiomics features for repeatability and reproducibility and identify robust features by evaluating the effects of biological and noise signals. A study [ 156 ] published in 2021 used a new method (which differs from embedded methods) for selecting robust features for predicting the mutation status of isocitrate dehydrogenase 1/2 (IDH1/2) in glioma.…”

Cancer radiomic feature variations due to reconstruction kernel choice and integral tube current.

“…(Haarburger et al, 2020; Ligero et al, 2021b, 2021a; Reiazi et al, 2021; Zhao et al, 2014). Other studies focus on the identification of radiomics features that are robust to batch effects, so they can be included in the radiomics signatures using the concordance and the correlations coefficient or the intraclass correlations coefficient (Carles et al, 2021; Lu et al, 2021). Finally, a body of work focuses on the development of batch effects correction methods and strategies for data harmonization and the minimization of acquisition-related variability (Cavinato et al, 2023; Eshaghzadeh Torbati et al, 2021; Horng et al, 2022; Lee et al, 2022; Ligero et al, 2021b; Mahon et al, 2020; Orlhac et al, 2022; Soliman et al, 2022).…”