Genomic and Glycolytic Entropy Are Reliable Radiogenomic Heterogeneity Biomarkers for Non-Small Cell Lung Cancer

Abstract

Radiogenomic heterogeneity features in 18F-fluorodeoxyglucose positron emission tomography (18F-FDG PET) have become popular in non-small cell lung cancer (NSCLC) research. However, the reliabilities of genomic heterogeneity features and of PET-based glycolytic features in different image matrix sizes have yet to be thoroughly tested. We conducted a prospective study with 46 NSCLC patients to assess the intra-class correlation coefficient (ICC) of different genomic heterogeneity features. We also tested the ICC of PET-based heterogeneity features from different image matrix sizes. The association of radiogenomic features with clinical data was also examined. The entropy-based genomic heterogeneity feature (ICC = 0.736) is more reliable than the median-based feature (ICC = −0.416). The PET-based glycolytic entropy was insensitive to image matrix size change (ICC = 0.958) and remained reliable in tumors with a metabolic volume of <10 mL (ICC = 0.894). The glycolytic entropy is also significantly associated with advanced cancer stages (p = 0.011). We conclude that the entropy-based radiogenomic features are reliable and may serve as ideal biomarkers for research and further clinical use for NSCLC.

The Ubiquitin–Proteasome System in Tumor Metabolism

“…The ATP yield rate is much faster in oxidative phosphorylation. The “Warburg effect” has been demonstrated in a variety of tumors, and radiofluorine-labeled glucose analogue uptake imaging based on positron emission tomography and 18F-fluorodeoxyglucose has been successfully used in clinical tumor diagnosis and staging, as well as to monitor response to treatment [ 31 ].…”

A comparison of 18F-FDG PET-based radiomics and deep learning in predicting regional lymph node metastasis in patients with resectable lung adenocarcinoma: a cross-scanner and temporal validation study

“…Additionally, although we can use the occlusion method to explain which part of the image activates the positive prediction [40], the underlying mechanism and pathophysiology remain elusive. Conversely, 18 F-FDG PET-based entropy correlates with genomic heterogeneity and is considered a biomarker of glycolytic heterogeneity [52,54]. As lung adenocarcinoma accumulates more heterogeneous clones that tend to metastasize [55], primary tumors with higher heterogeneity are more likely to reveal positive regional nodes.…”

“…However, not all exhibit high reliability [30,31,[50][51][52]. Nevertheless, the intensity and volumetric features we used, as well as entropy, are highly reliable in digital PET systems [52]. We thus used these features in our study.…”

“…Nevertheless, unlike DL, which demonstrated reduced predictive accuracy when applied to digital PET (94.7% and 88.2% in the analog and digital PET cohorts, respectively), the accuracy of entropy did not reduce after cross-scanner validation (62.6% and 64.7% in the analog and digital PET cohorts, respectively). Furthermore, the reliability of entropy has been well established [30,31,50–53]. Therefore, the stability of the cross-scanner diagnostic performance may suggest that the predictive performance of entropy may be more reproducible than that of DL.…”

“…Third, in addition to the radiomic features used in our study, many other radiomic features are available. However, not all exhibit high reliability [30,31,50–52]. Nevertheless, the intensity and volumetric features we used, as well as entropy, are highly reliable in digital PET systems [52].…”