Lung cancer is the third most common cancer and the leading cause of cancer-related death in the United States, with most subtypes of lung cancers having poor prognosis and low survival rates.1,2 Diagnostic imaging with CT has become a mainstay during oncologic treatment. However, its inability to convey prognostic information has led to the emergence of radiomics, or quantitative imaging. Radiomics involves imaging biomarkers that can decode information about tumor phenotype and intratumoral heterogeneity imbedded within conventional imaging. It also has the potential to modernize oncologic management by providing prognostic analyses relative to genomic mutational profile, histologic subtype, and risk of distant metastasis.1-6
Recent work by Khorrami et al.7 retrospectively outlined the predictive value of intra- and peritumoral radiomic features on baseline CT exams in 125 patients with NSCLC treated with pemetrexed-based platinum doublet chemotherapy. In a series of well-documented analyses, they concluded that radiomics features both in and around the segmented tumor on CT are predictive of response to chemotherapy and associated with time to progression and OS.
The study addressed reproducibility, a recent challenge in machine learning analyses of high-throughput quantitative image analysis, by using test–retest RIDER and measuring the intraclass correlation coefficient (ICC) between them. This approach tests the reliability of the features (ICC > 0.8) and reduces the high dimensionality of radiomics (76%), which is exacerbated by the intersecting of ICC among readers (34%). Moreover, the authors considered higher Dice similarity coefficients (> 80) in order to tackle the problem of inter-reader variability. Applying a LASSO Cox regression model mitigated the chance of overfitting resulting from increasing the complexity of the model due to an additive penalty term. Yet the analyses might suffer from the selection of tuning parameters, which was addressed by using a 100-fold cross-validation that included minimal criteria.
There have been several studies examining the radiomic signatures in the peritumoral zone of various cancers8-11, however, the authors’ novel contribution to the literature was to use clinically validated biomarkers to specifically identify patients with NSCLC who would benefit from platinum-based chemotherapy agents. This study is particularly important with regard to current treatment of NSCLC. The authors explained that although platinum-based chemotherapy is one of the first-line treatments of advanced-stage NSCLC with no actionable mutations, the outcomes and survival rates are quite modest, and the objective response rate to this regimen as initial treatment is only around 24% to 31%. Therefore, it can be inferred that identifying patients who may not benefit from this regimen would spare these patients potentially toxic side effects with little to no benefit from therapy. By the same token, the authors inferred that capturing tumor heterogeneity can potentially identify patients at elevated risk for recurrence who may need closer follow-up.
It is known that tumor heterogeneity is a predictor of survival in NSCLC. The reason for this is unclear, but it may reflect genomic heterogeneity or clonal dominance. The authors went one step further by discussing how peritumoral texture features that can be analyzed with radiomics may reflect the microenvironment of tumors. This microenvironment potentially represents increased fibrotic content in chemotherapy-compliant tumors, tumor hypoxia, and tumor-infiltrating lymphocytes as well as tumor-associated macrophages, all of which may affect the efficacy of chemotherapy.
However, the authors stated the limitation of their study concerning the morphologic and molecular basis analysis for the observed peritumoral radiomic features and attributed it to their relatively small sample size. In addition, their study was a single-institution design, which might affect generalizability of the classifier. The authors noted the influence of CT acquisition parameters on radiomic feature analysis, acknowledging the fact that several CT acquisition parameters, particularly slice thickness12 and pre- and post-processing algorithms,13 introduced variability in segmentations, which they hope to address in future work. An additional potential limitation was the differences in manual segmentations, which has been known to introduce variability in radiomic feature analysis; however, the authors addressed this issue well by using three readers and comparing segmentation accuracy.
Ultimately, this original study highlighted several important points regarding the application of radiomics in prognostic analyses of patients with NSCLC. It emphasized not only intratumoral radiomic features but also surrounding tumor microenvironments, to enable prediction of who will respond to pemetrexed-based chemotherapy and who will not. The study was also able to use baseline CT–derived tumor heterogeneity as a prediction of time-to-progression and OS in patients with NSCLC. A brief note was also made of the potential impact that peritumoral texture feature analysis has on immunotherapy, which is currently included in the treatment plan in many patients with NSCLC as first line therapy, either alone or in conjunction with chemotherapy. Further validation would be needed to support this point, but it is an interesting topic in the era of immune checkpoint inhibitors. Although the authors acknowledged that additional large-scale multi-institution validation must be performed before clinical deployment of the model, they have provided important feasibility data on the potential prognostic ability of using radiomic feature analysis for prediction of therapy response to pemetrexed-based chemotherapy.
- Huang Q, Lu L, Dercle L, et al. Interobserver variability in tumor contouring affects the use of radiomics to predict mutational status. J Med Imaging (Bellingham). 2018; 5:011005.
- Balagurunathan Y, Kumar V, Gu Y, et al. Test-retest reproducibility analysis of lung CT image features. J Digit Imaging. 2014;27:805-823.
- Wu W, Parmar C, Grossmann P, et al. Exploratory Study to Identify Radiomics Classifiers for Lung Cancer Histology. Front Oncol. 2016;6:71.
- Fried DV, Tucker SL, Zhou S, et al. Prognostic value and reproducibility of pretreatment CT texture features in stage III non-small cell lung cancer. Int J Radiat Oncol Biol Phys. 2014;90:834-842
- Aerts HJ, Grossmann P, Tan Y, et al. Defining a Radiomic Response Phenotype: A Pilot Study using targeted therapy in NSCLC. Sci Rep. 2016;6:33860.
- Coroller TP, Grossmann P, Hou Y, et al. CT-based radiomic signature predicts distant metastasis in lung adenocarcinoma. Radiother Oncol. 2015;114:345-350.
- Khorrami M, Khunger M, Zagouras A, et al. Combination of Peri- and Intratumoral Radiomic Features on Baseline CT Scans Predicts Response to Chemotherapy in Lung Adenocarcinoma. Radiol Artif Intell. 2019;1:e180012.
- Dou TH, Coroller TP, van Griethuysen JJM, Mak RH, Aerts HJWL. Peritumoral radiomics features predict distant metastasis in locally advanced NSCLC. PLoS One. 2018;13:e0206108.
- Braman NM, Etesami M, Prasanna P, et al. Intratumoral and peritumoral radiomics for the pretreatment prediction of pathological complete response to neoadjuvant chemotherapy based on breast DCE-MRI. Breast Cancer Res. 2017;19:57.
- Akinci D’Antonoli T, Farchione A, Lenkowicz J, et al. CT Radiomics Signature of Tumor and Peritumoral Lung Parenchyma to Predict Nonsmall Cell Lung Cancer Postsurgical Recurrence Risk. Acad Radiol. 2020;27(4):497-507.
- Khorrami M, Jain P, Bera K, et al. Predicting pathologic response to neoadjuvant chemoradiation in resectable stage III non-small cell lung cancer patients using computed tomography radiomic features. Lung Cancer. 2019;135:1-9.
- Zhao B, Tan Y, Tsai WY, et al. Reproducibility of radiomics for deciphering tumor phenotype with imaging. Sci Rep. 2016;6:23428.
- Traverso A, Wee L, Dekker A, Gillies R. Repeatability and Reproducibility of Radiomic Features: A Systematic Review. Int J Radiat Oncol Biol Phys. 2018;102:1143-1158.