Radiomic phenotypes of oligometastatic non-small cell lung cancer

Introduction

More than half of patients with non-small cell lung cancer (NSCLC) are diagnosed with metastatic disease, which portends a 5-year relative overall survival (OS) of just 27% (1). In the past decade, new and effective interventions have arisen in clinical practice, including targeted systemic therapies indicated by genomic aberrations (2,3) and immunotherapy (4). However, since most patients develop resistance to systemic therapy, the integration of local ablative therapy (LAT) has gained increased interest, particularly in the setting of more indolent metastatic disease (5-10). Furthermore, lesions that persist or progress despite systemic therapy can be effectively ablated and eliminated using stereotactic body radiotherapy (SBRT): this LAT-based approach can offer a reprieve for patients enduring long courses of systemic therapy, as recommended in European Organisation for Research and Treatment of Cancer (EORTC) and European Society for Radiotherapy and Oncology (ESTRO) guidelines (11).

Selection for LAT has typically been based on the number of metastatic lesions, and there are no established imaging or molecular biomarkers. This approach is both incomplete and often arbitrary. Literature reviews have suggested that certain genomic aberrations, such as KRAS mutations and unchecked NOTCH pathway signaling, are associated with greater metastatic potential (12,13). A complementary approach to enhancing the quality of selecting patients for aggressive treatment is to integrate analysis of diagnostic imaging and tissue prior to treatment. In this context, fluorodeoxyglucose (FDG) positron emission tomography (PET) and computed tomography (CT) imaging studies are routinely acquired for disease staging. Radiomic features, or quantitative features describing shapes and patterns not always visible to the human eye (14), have shown promising preclinical results in NSCLC for prognostication and prediction of response to therapy using CT (15-18) and PET (19-24).

In this study, we investigated whether and which radiomic features of the primary lesion on positron emission tomography/computed tomography (PET/CT) imaging of metastatic NSCLC are associated with extent of metastatic disease and indolent disease biology via OS. We hypothesize that larger and more infiltrative tumors will correlate with greater extent of metastatic disease and shorter OS. We present this article in accordance with the STROBE reporting checklist (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-159/rc).

Methods

Inclusion criteria and cohort assignment

In this retrospective study, we extended a previously curated cohort (25) to yield a cohort of 167 patients diagnosed 2016–2020 (Table S1) with treatment-naïve, metastatic (any number of metastases) NSCLC who underwent pre-treatment FDG PET/CT imaging with available Digital Imaging and Communications in Medicine (DICOM) images and the majority of whom (n=153; Figure S1) also underwent molecular profiling of their disease through one of two methods: MSK-IMPACT (Integrated Mutation Profiling of Actionable Cancer Targets, tissue) or MSK-ACCESS (Analysis of Circulating cfDNA to Examine Somatic Status, blood), though molecular profiling was not an inclusion criterion (see Figure S1 for further details). The experimental protocol was approved by the Memorial Sloan Kettering Institutional Review Board (No. 17-014). All methods were performed in accordance with the relevant guidelines and regulations. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Informed consent was waived for this retrospective study.

Specifically, all patients presented with biopsy-proven NSCLC with a clinical diagnosis of metastatic disease and at least one segmentable lesion in the lung parenchyma for even comparison of radiomic features from the primary site of disease. Patients who received systemic therapy or underwent resection prior to imaging were not eligible. Both smokers and non-smokers were eligible, as were all races and ages. Patients were randomly allocated to the training/validation or test cohorts (80% and 20%, respectively). No model adjustments were made based on the test set; PET images with avidities reported in units other than becquerels per milliliter or corrupted files were excluded from the analysis (Figure S1). Treatment was not standardized or recorded and was at the treating physician’s discretion.

Genomics

MSK-IMPACT and MSK-ACCESS are sequencing platforms that have been well-described in prior publications (26,27). For the purposes of this study, we analyzed sequencing results that were reported per our standard pipeline and mechanisms, as described in prior literature (https://github.com/mskcc/Innovation-IMPACT-Pipeline).

With these results, we then identified variants annotated as pathogenic by OncoKB (28-30) in genes whose function has been found to associate with indolent oligometastatic disease in previous studies (12,13), specifically KRAS, SMAD4, NOTCH pathway, FGFR3, ERBB2, ALK, KMT2B, and PIK3CB. Programmed death ligand 1 (PD-L1) positivity as defined as >1% tumor expression on immunohistochemistry, was also included as a feature. The number of metastases was also considered for survival analyses.

Segmentation

To establish the number of metastatic lesions per patient, two readers delineated the extent of disease by segmenting at the voxel level for all patients using MIM Software (version 6.9.7; MIM Software Inc., Beachwood, OH, USA). All sites of disease >1 cm in axial diameter on cross-sectional imaging and with pathologic avidity on PET were included for the purposes of analysis. Both PET and CT imaging modalities were used to identify lesions. The imaging was acquired across multiple scanners at slice thicknesses ranging from 2.5–5.0 mm on equipment manufactured by Siemens, GE, and Philips acquired between 2016 and 2020, primarily at our institution (60%; Table S1).

Determining lesion numbers for analysis

Each lesion was labeled by anatomic site and categorized accordingly: lung, bone, nodes (unspecified), thoracic nodes, pleura, adrenal, hepatic, soft tissue, abdominal nodes, supraclavicular nodes, axillary nodes, cervical nodes, thyroid, submandibular gland, heart, renal, spinal cord intramedullary, and pancreas. After segmentation, by means of connected components analysis, we tallied the number of individual lesions for each patient. Consistent with prior published randomized studies, any lesion in each nodal group was counted as one single “lesion”, though as noted above all patients also had distant disease (31). This automated counting of metastatic burden correlated well with manual assessment by a physician (Table S2).

Radiomic feature extraction and selection

We converted all DICOM CT series to Hounsfield units and concatenated them into volumetric images. We rigidly coregistered the DICOM PET series with the CT series and adjusted PET voxels to the same size as CT voxels. We then converted PET avidities from Becquerels/mL to standard uptake values (SUV) based on body weight and adjusted for decay between radiopharmaceutical administration and image acquisition using 3D Slicer’s (32) PET DICOM extension (https://github.com/QIICR/Slicer-PETDICOMExtension). We also converted DICOM RT structures to volumetric voxel-wise masks using SimpleITK (33).

Using PyRadiomics (34), we binned images at native resolution with bin sizes of 10 Hounsfield units (HU) and 1 normalized SUV units (for CT and PET, respectively). Only features from the largest intraparenchymal lung lesion were extracted to avoid confounding with extent of disease. For CT images, features were extracted from images in the lung window (range, −1,350 to 150 HU). We extracted features in 3D from original images, including features from the gray level size zone matrix (35), yielding a 48-dimensional representation for each imaging modality. Features were scaled to have a mean of zero and a standard deviation of one for the training set.

To reduce the influence of confounders, we assessed the association of each feature with acquisition year, vendor, slice thickness, and whether the image was acquired at our institution or externally. The Mann-Whitney U test was used for dichotomous variables, and the Kruskal-Wallis test was used for multi-class classification. Any features with uncorrected P values below 0.05 were discarded; features derived from each imaging modality were tested separately (Table S3). Features with >0.90 Pearson correlation with another feature were removed.

Statistical analysis

To select features associated with extent of metastatic disease (number of separate metastatic lesions), the Pearson correlation was used to compare the normalized values of each feature to the raw number of separate lesions (all of which were two or greater) within the training/validation set. After Benjamini-Hochberg correction for multiple testing, the features significantly associated (q≤0.05) with the number of lesions were selected for the model. To select features associated with OS, we used the joint training/validation cohort to fit univariate Cox Proportional Hazards models using the Python Lifelines software package with default settings, and all features with a log hazard ratio (HR) differing significantly from zero with 95% confidence interval (CI) (after Benjamini-Hochberg correction for multiple testing) were selected for downstream modeling. Features derived from computed and positron emission tomography were considered separately. The same feature selection process was used for clinicogenomic features, but correction for multiple testing was not used for clinicogenomic features (P≤0.05) because of the low number of features tested.

Modeling extent of metastatic disease and OS

To model the extent of metastatic disease, the selected features were used for ordinary least squares on the training set, then validated on the validation and test cohorts. Models were assessed by the Pearson correlation of the inferred score with the extent of metastatic disease.

Given the smaller numbers with annotated OS, the Cox proportional hazards model was fit on the combined training and validation sets and tested on the test set. If multiple radiomic features were selected, features with most uncertain significance were iteratively removed until all log HRs differed from zero with 95% CI. The partial hazards were used to stratify patients, and a cutpoint for high- and low-risk groups was selected based on the training/validation cohort (selecting the cutpoint yielding two risk groups with the greatest confidence in their difference, permitting as few as 15% in one risk group or the other). The same cutpoint was used on the test set. Log-rank tests were used to assess significance.

Results

Cohort characteristics

We included 167 patients with stage IV NSCLC who met the eligibility criteria above (Table S1; Figure 1A; Table 1). These patients underwent PET/CT imaging (Figure 1B), physician segmentation of the entire disease burden (Figure 1C), and extraction of 48 radiomic features (Figure 1D). The mean number of lesions was 11.5, the median was seven, and the range was 2–82 (Figure 1E). The patients were divided into a training cohort (79%, n=132) and a validation cohort (21%, n=35) without stratification. The most common sites for lesions were lung (n=167), bone (n=98), and thoracic nodes (n=74; Table S4). Sex distribution was evenly balanced in the training (50% men) and validation cohorts (46% men; Table 1). Of patients with tested genomic alterations, the most common alterations were observed in KRAS (31%) and SMAD4 (9%) (Figure 1F). Median follow-up was 16.3 months (standard deviation 15.9 months) and 25.4 months (standard deviation 12.2 months) for the training/validation (n=66) and test (n=17) sets, respectively. In the training/validation cohort, 26 cases were censored. In the test cohort, 4 cases were censored. PET/CT images were acquired from 2016 to 2020 (Table S1).

Figure 1 Schematic overview of study. (A) Patients with metastatic NSCLC underwent (B) FDG PET/CT imaging, (C) subsequent segmentation of total disease burden, and (D) standardized radiomic feature extraction. (E) Number of lesions for each case. (F) Oncoprint for the cohort. CT, computed tomography; FDG, fluorodeoxyglucose; NLP, natural language processing; NSCLC, non-small cell lung cancer; PD-L1, programmed death-ligand 1; PET, positron emission tomography.

Primary lesions for oligometastatic cases were smaller and more spherical

Using PET/CT images, we manually contoured the entire burden of disease and extracted radiomic features unconfounded by scanner parameters such as slice thickness, manufacturer, institution, and acquisition year (see Methods). Using MSK-IMPACT and MSK-ACCESS sequencing results, we assessed the mutational status of genes previously found to be associated with latent disease phenotype in NSCLC (12,13). We found that, using only OncoKB-annotated variants, the number of lesions did not differ significantly with lesions to these genes (Table 2).

When assessing radiomic features of the primary lung lesion, we found that two features were associated with the extent of metastatic disease after correction for multiple testing (Figure 2A). The surface area-to-volume ratio (SVR) and the inverse of the flatness were higher (r=0.24; q=0.04) and lower (r=−0.23; q=0.04), respectively, for cases with more extensive disease. These two features did not remain significant upon multivariate modeling to infer number of metastatic lesions (inverse flatness coef. =−2.08; 95% CI: −4.789 to 0.633, SVR coef. =2.21; 95% CI: −0.497 to 4.925; Figure 2B). However, the two features exhibited only limited multicollinearity; with a Pearson correlation of −0.27 (Figure 2C). The model’s inferred number of metastases correlated modestly with the true number for the training (r=0.30; P<0.001) and test (r=0.24; P=0.17) sets (Figure 2D). Examples of cases with high- and low-SVR are depicted in Figure 2E,2F, respectively. SVR is not unitless, and we show it depends partly on lesion volume (Pearson’s r=−0.45; P<0.001) in Figure S2. Examples of cases with high- and low-inverse flatness are depicted in Figure 2G,2H, respectively. No tested features derived from the signal intensity of CT or PET were associated with the extent of metastatic disease (Figure 2A; Figure S3).

Figure 2 Radiomic correlates of extent of disease. (A) Association of CT-derived radiomic features with number of metastatic lesions; (B) forest plot for multivariate regression of number of metastatic lesions depicting 95% confidence interval for each coefficient; (C) Pearson’s correlation between the two variables; (D) model correlation with disease burden in the training/validation and test sets. Example features for cases with (E) high SVR, (F) low SVR, (G) high inverse flatness, and (H) low inverse flatness. CT, computed tomography; Flat.−1, inverse flatness; GLE, gray level emphasis; SVR, surface area-to-volume ratio.

Higher gray level variance of primary lesions portends shorter OS

By selecting features using univariate Cox proportional hazards models from the same set of genomic aberrations along with clinical variables, we found that binary PD-L1 receptor status (>1% tumor expression) (HR =0.45, 95% CI: 0.22–0.94, P=0.03) and number of metastases (HR =1.02, 95% CI: 1.002–1.03, P=0.03) were significantly associated with OS (test c=0.63, 95% CI: 0.41–0.86; Table 3). We performed the same feature selection procedure on the unconfounded radiomic feature set (Figure 3A), identifying the gray level variance based on the gray level size zone matrix as significantly associated with OS (Figure 3B; HR =1.84, 95% CI: 1.32–2.56, q=0.004; test c=0.74, 95% CI: 0.52–0.90). Using early fusion (36), we integrated the clinicogenomic and radiomic features and found that the concordance index did not differ significantly from the unimodal values for the test set: c=0.69 (Table S5; Figure S4; 95% CI: 0.46–0.89). For the radiomic model, we selected a cutpoint based on the maximal separation of high- and low-risk groups in the training set (see Methods) and found that these groups exhibited median survival times in the training set of 12.6 and >50 months (Figure 3C; P=0.001) and in the test set of 13.8 and 32.7 months (Figure 3D; P=0.065). Upon visualizing lesions with low (Figure 3E) and high (Figure 3F) gray level variance, we noted that the contoured lesions with lower gray level variance contained less low-density airspace (Figure 3G). Empirically, the value of this feature was associated with the percent of voxels with intensities below zero Hounsfield units (r=0.47, P<0.001; Figure S5). This suggests increased complexity of the boundary between tumor voxels and the surrounding lower-density airspace, and thus more infiltrative disease.

Figure 3 Radiomic correlates of overall survival. (A) Association of CT-derived radiomic features with overall survival; (B) forest plot for multivariate regression of number of metastatic lesions depicting 95% confidence interval for each coefficient; (C) Kaplan-Meier analysis of high- and low-risk groups in the training/validation sets; (D) Kaplan-Meier analysis of high- and low-risk groups in the test set; (E) case with low gray level variance; (F) case with high gray level variance; (G) plots of voxel values for cases in (E, bottom) and (F, top). CT, computed tomography; GLE, gray level emphasis; SVR, surface area-to-volume ratio.

Discussion

From this study of patients with metastatic NSCLC, we draw two conclusions: first, we have identified that flatter primary lesions with greater ratios of surface area to volume are more likely to be associated with a greater extent of metastatic disease than more spherical primary lesions. Second, we have identified that primary lesions with greater gray level variance—driven by inclusion of lower-density pulmonary airspaces—are associated with shorter OS.

Regarding the radiomic features associated with extent of disease, SVR has been established in other radiomic studies of NSCLC as associated with likelihood of malignancy of nodules (37-39). Our finding of its association with a greater extent of metastatic disease parallels this association on the continuum of disease aggressiveness. On visual inspection and empirical testing, we further observed its inverse correlation with absolute volume given that it is not a dimensionless feature. Analogously, inverse flatness of the primary lesion was lower for cases with more extensive metastatic disease, suggesting that smaller, well-delineated, spherical lesions are associated with more favorable risk disease. Such shape-based radiomic features have been borne out as especially reproducible across multiple readers for NSCLC (40) with both flatness and surface area-to-volume ratio exhibiting intra-class correlation coefficients of greater than 0.8 and low sensitivity to confounders. These shape-based features are derived from the primary tumor alone, which are easy to inspect or segment on PET/CT imaging and may assist clinicians in distinguishing more aggressive disease with a greater risk of extensive metastatic spread from oligometastatic disease more likely to benefit from LAT. Previous work has established SUV_max as a correlate of occult nodal disease (19,20), but our analysis did not identify it as also associated with overall extent of metastatic disease. Primary tumor size is also well established to be associated with aggressive disease (41,42), but in our analysis, raw primary tumor volume was associated with the acquisition year and thus not considered during model selection. However, surface area-to-volume ratio feature is partly determined by total volume. Regarding our molecular analysis, we hypothesize that the lack of significant associations between genomic aberrations previously associated with oligometastatic state and extent of metastatic disease in our cohort is at least partially due to small sample size.

For our prognostic investigation, we identified PD-L1 negativity and greater number of metastases as associated with shorter OS in accordance with established literature (43-45). The gray level variance of the primary lesion stratified patients by OS at a level comparable to these established biomarkers: greater gray level variance was associated with shorter OS. We further showed that this higher gray level variance was driven largely by inclusion of voxel values below zero Hounsfield units. This suggests that lower-density tumoral regions and inclusion of airspace due to higher-complexity tumor borders portend worse prognosis than dense, well-defined primary tumors. Gray level variance was not confounded by any variables we tested, but its value varies by reader, with an intra-class correlation coefficient of approximately 0.2 in one study (40). However, this analysis involved rough segmentations as a comparison to intentionally accurate segmentations, while an analysis comparing only careful contours in lung found that all texture-based features, including those derived from the gray level size zone matrix, exhibited intra-class correlation coefficient values of 0.91±0.11 (34). Previous radiomic signatures have separated patients by tumoral and nodal stage but not by metastatic stage (46) or extent of disease. Hence, we have identified a radiomic signature for patients exclusively with metastatic NSCLC, which extends such previous studies that included patients at varied stages of disease (17). Further work with automatic segmentation of the nodules could enhance the robustness of this feature to variations in segmentation and further establish whether inclusion of lower-density lung parenchyma adjacent to the tumor, due to a complex tumor-pulmonary interface or lower-density tumor itself is driving this association.

The main limitation of our study is the sample size. We took a focused approach to radiomic modeling, but a larger, multi-institutional cohort could enable the testing of more features (36). Another limitation is the dependence of radiomic feature values on segmentation (47). Multi-reader segmentation could ameliorate this by deriving consensus segmentations based on each reader’s contours. To address this in our study, we used image biomarker standardisation initiative (IBSI)-defined radiomic features with demonstrably strong reproducibility across scanners and acquisition protocols (14), and experienced radiation oncologists reviewed and edited contours as needed. Future work could incorporate advanced imaging modalities: DeSouza and Tempany also posit an association between angiogenic potential and polymetastatic disease (22) based on evidence in other disease histologies, but this requires perfusion imaging and has not yet been empirically tested in NSCLC.

LAT in NSCLC offers an attractive treatment paradigm to complement systemic therapy, especially in the setting of oligoprogressive disease (11). Recent investigations in the field have empirically established the association between larger clinical target volume (CTV) size and post-therapy lymphopenia/leukopenia, which associates with diminished response to checkpoint blockade (48). Ongoing research seeks to establish the optimal dosing, CTV definitions, interplay between SBRT and the immune system (49), and definition of oligometastatic disease to which SBRT could be effectively applied. Radiomic biomarkers ought to be investigated as indicators of disease to which SBRT could be applied.

Conclusions

In summary, we conducted a multimodal analysis of a unique dataset of 167 patients with metastatic NSCLC. We found that polymetastatic disease characteristically exhibited flatter primary lesions with greater surface area-to-volume ratios, thereby suggesting quantitative analysis of the primary lesion as a noninvasive approach to assess disease aggressiveness, which in turn can be used to stratify patients for LAT. In addition, we identified greater gray level variance as a strong correlate of OS on par with previously identified clinicogenomic features. Few if any studies to date have examined radiomic features associated with extent of disease, a crucial indicator of a patient’s likelihood to benefit from LAT. Taken together, our findings represent progress toward better selecting patients for LAT upon further validation. Such models are expected to improve outcomes and individualized therapies for patients with metastatic disease.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STROBE reporting checklist. Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-159/rc

Data Sharing Statement: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-159/dss

Peer Review File: Available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-159/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://tlcr.amegroups.com/article/view/10.21037/tlcr-2025-159/coif). K.M.B. is a consultant from Monograph Capital, and received meeting support from La Fondazione IRCCS Istituto Nazionale dei Tumori; honorarium from Japanese Society of Obstetrics and Gynecology. A.R. reports funding support from AstraZeneca, Merck, Boehringer Ingelheim, Pfizer, Varian Medical Systems to their institution; consulting fees from AstraZeneca, Merck, MoreHealth; honoraria from Boehringer Ingelheim; travel support from AstraZeneca; participation as a board member of Merck; he is also the Vice President of ITMIG, on the Board of Directors of IMIG and an oral board examiner of ABR. N.S. reports research funding from AstraZeneca, Novartis and Amgen (to institution). H.Y. received consulting fees from AstraZeneca, Janssen, BMS, Daiichi, Ipsen, Orion, Abbvie, Pfizer, Amgen, Takeda and Taiho. H.Y. is on the DSMB for a Janssen study and a Mythic Therapeutics study. B.T.L. reports funding from National Institutes of Health, Amgen, AstraZeneca, Bolt Biotherapeutics, Daiichi Sankyo, Genentech, Jiangsu Hengrui Pharmaceuticals, Lily, Nuvalent, Revolution Medicines; academic travel support from Amgen; patent in Memorial Sloan Kettering Cancer Center; and owns stock of AstraZeneca. B.T.L. is a senior fellow of Asia Society Policy Institute, co-founder of Bloomberg New Economy International Cancer Coalition and Cure4Cancer and an employee of AstraZeneca. J.M.I. received consulting fees from AstraZeneca, Merck as Advisory Board Member. D.R.G. reports receiving funding from Johnson & Johnson, Amgen, AstraZeneca, Varian; consulting fees from Johnson & Johnson, AstraZeneca, Regeneron, Medtronic, Olympus, Grail; and other payments from MedLearning Group, CME Clinical Group, Varian. The other authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The experimental protocol was approved by the Memorial Sloan Kettering Institutional Review Board (No. 17-014). This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. Informed consent was waived for this retrospective study.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Cancer of the Lung and Bronchus - Cancer Stat Facts. SEER. [cited 2024 Sep 28]. Available online: https://seer.cancer.gov/statfacts/html/lungb.html
2. Dong YU, Ren W, Qi J, et al. EGFR, ALK, RET, KRAS and BRAF alterations in never-smokers with non-small cell lung cancer. Oncol Lett 2016;11:2371-8. [Crossref] [PubMed]
3. Rothschild SI. Targeted Therapies in Non-Small Cell Lung Cancer-Beyond EGFR and ALK. Cancers (Basel) 2015;7:930-49. [Crossref] [PubMed]
4. Grant MJ, Herbst RS, Goldberg SB. Selecting the optimal immunotherapy regimen in driver-negative metastatic NSCLC. Nat Rev Clin Oncol 2021;18:625-44. [Crossref] [PubMed]
5. Tsai CJ, Yang JT, Guttmann DM, et al. Final analysis of consolidative use of radiotherapy to block (CURB) oligoprogression trial - A randomized study of stereotactic body radiotherapy for oligoprogressive metastatic lung and breast cancers. Int J Radiat Oncol Biol Phys 2022;114:1061.
6. Gomez DR, Yang TJ, Tsai CJ. Emerging Paradigm of Consolidative Thoracic Radiotherapy in Oligometastatic NSCLC. Semin Radiat Oncol 2021;31:120-3. [Crossref] [PubMed]
7. Iyengar P, All S, Berry MF, et al. Treatment of Oligometastatic Non-Small Cell Lung Cancer: An ASTRO/ESTRO Clinical Practice Guideline. Pract Radiat Oncol 2023;13:393-412. [Crossref] [PubMed]
8. Hubbeling H, Choudhury N, Flynn J, et al. Outcomes With Local Therapy and Tyrosine Kinase Inhibition in Patients With ALK/ROS1/RET-Rearranged Lung Cancers. JCO Precis Oncol 2022;6:e2200024. [Crossref] [PubMed]
9. Yu HA, Sima CS, Huang J, et al. Local therapy with continued EGFR tyrosine kinase inhibitor therapy as a treatment strategy in EGFR-mutant advanced lung cancers that have developed acquired resistance to EGFR tyrosine kinase inhibitors. J Thorac Oncol 2013;8:346-51. [Crossref] [PubMed]
10. Palma DA, Olson R, Harrow S, et al. Stereotactic Ablative Radiotherapy for the Comprehensive Treatment of Oligometastatic Cancers: Long-Term Results of the SABR-COMET Phase II Randomized Trial. J Clin Oncol 2020;38:2830-8. [Crossref] [PubMed]
11. Guckenberger M, Lievens Y, Bouma AB, et al. Characterisation and classification of oligometastatic disease: a European Society for Radiotherapy and Oncology and European Organisation for Research and Treatment of Cancer consensus recommendation. Lancet Oncol 2020;21:e18-28. [Crossref] [PubMed]
12. Liao R, Chen K, Li J, et al. Genomic Characteristics and the Potential Clinical Implications in Oligometastatic Non-Small Cell Lung Cancer. Cancer Res Treat 2023;55:814-31. [Crossref] [PubMed]
13. Rashdan S, Iyengar P, Minna JD, et al. Narrative review: molecular and genetic profiling of oligometastatic non-small cell lung cancer. Transl Lung Cancer Res 2021;10:3351-68. [Crossref] [PubMed]
14. Zwanenburg A, Vallières M, Abdalah MA, et al. The Image Biomarker Standardization Initiative: Standardized Quantitative Radiomics for High-Throughput Image-based Phenotyping. Radiology 2020;295:328-38. [Crossref] [PubMed]
15. Coroller TP, Agrawal V, Huynh E, et al. Radiomic-Based Pathological Response Prediction from Primary Tumors and Lymph Nodes in NSCLC. J Thorac Oncol 2017;12:467-76. [Crossref] [PubMed]
16. Thawani R, Mustafa SA. The future of radiomics in lung cancer. Lancet Digit Health 2020;2:e103. [Crossref] [PubMed]
17. Hosny A, Parmar C, Coroller TP, et al. Deep learning for lung cancer prognostication: A retrospective multi-cohort radiomics study. PLoS Med 2018;15:e1002711. [Crossref] [PubMed]
18. Thawani R, McLane M, Beig N, et al. Radiomics and radiogenomics in lung cancer: A review for the clinician. Lung Cancer 2018;115:34-41. [Crossref] [PubMed]
19. Li L, Ren S, Zhang Y, et al. Risk factors for predicting the occult nodal metastasis in T1-2N0M0 NSCLC patients staged by PET/CT: potential value in the clinic. Lung Cancer 2013;81:213-7. [Crossref] [PubMed]
20. Kaseda K, Asakura K, Kazama A, et al. Risk Factors for Predicting Occult Lymph Node Metastasis in Patients with Clinical Stage I Non-small Cell Lung Cancer Staged by Integrated Fluorodeoxyglucose Positron Emission Tomography/Computed Tomography. World J Surg 2016;40:2976-83. [Crossref] [PubMed]
21. Nair VJ, MacRae R, Sirisegaram A, et al. Pretreatment [18F]-fluoro-2-deoxy-glucose positron emission tomography maximum standardized uptake value as predictor of distant metastasis in early-stage non-small cell lung cancer treated with definitive radiation therapy: rethinking the role of positron emission tomography in personalizing treatment based on risk status. Int J Radiat Oncol Biol Phys 2014;88:312-8.
22. deSouza NM, Tempany CM. A risk-based approach to identifying oligometastatic disease on imaging. Int J Cancer 2019;144:422-30. [Crossref] [PubMed]
23. Shin S, Pak K, Park DY, et al. Tumor Heterogeneity Assessed by 18F-FDG PET/CT Is Not Significantly Associated with Nodal Metastasis in Breast Cancer Patients. Oncol Res Treat 2016;39:61-6. [Crossref] [PubMed]
24. Pass HI, Carbone DP, Johnson DH, et al. Principles and Practice of Lung Cancer: The Official Reference Text of the International Association for the Study of Lung Cancer (IASLC). Philadelphia: Lippincott Williams & Wilkins; 2012.
25. Haseltine JM, Offin M, Flynn JR, et al. Tumor volume as a predictor of cell free DNA mutation detection in advanced non-small cell lung cancer. Transl Lung Cancer Res 2022;11:1578-90. [Crossref] [PubMed]
26. Cheng DT, Mitchell TN, Zehir A, et al. Memorial Sloan Kettering-Integrated Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT): A Hybridization Capture-Based Next-Generation Sequencing Clinical Assay for Solid Tumor Molecular Oncology. J Mol Diagn 2015;17:251-64. [Crossref] [PubMed]
27. Rose Brannon A, Jayakumaran G, Diosdado M, et al. Enhanced specificity of clinical high-sensitivity tumor mutation profiling in cell-free DNA via paired normal sequencing using MSK-ACCESS. Nat Commun 2021;12:3770. [Crossref] [PubMed]
28. Chakravarty D, Gao J, Phillips SM, et al. OncoKB: A Precision Oncology Knowledge Base. JCO Precis Oncol 2017;2017:PO.17.00011.
29. Chang MT, Bhattarai TS, Schram AM, et al. Accelerating Discovery of Functional Mutant Alleles in Cancer. Cancer Discov 2018;8:174-83. [Crossref] [PubMed]
30. Chang MT, Asthana S, Gao SP, et al. Identifying recurrent mutations in cancer reveals widespread lineage diversity and mutational specificity. Nat Biotechnol 2016;34:155-63. [Crossref] [PubMed]
31. Gomez DR, Blumenschein GR Jr, Lee JJ, et al. Local consolidative therapy versus maintenance therapy or observation for patients with oligometastatic non-small-cell lung cancer without progression after first-line systemic therapy: a multicentre, randomised, controlled, phase 2 study. Lancet Oncol 2016;17:1672-82. [Crossref] [PubMed]
32. Fedorov A, Beichel R, Kalpathy-Cramer J, et al. 3D Slicer as an image computing platform for the Quantitative Imaging Network. Magn Reson Imaging 2012;30:1323-41. [Crossref] [PubMed]
33. Lowekamp BC, Chen DT, Ibáñez L, et al. The Design of SimpleITK. Front Neuroinform 2013;7:45. [Crossref] [PubMed]
34. van Griethuysen JJM, Fedorov A, Parmar C, et al. Computational Radiomics System to Decode the Radiographic Phenotype. Cancer Res 2017;77:e104-7. [Crossref] [PubMed]
35. Thibault G, Fertil B, Navarro C, et al. Shape and texture indexes application to cell nuclei classification. Int J Pattern Recognit Artif Intell 2013;27:1357002.
36. Boehm KM, Khosravi P, Vanguri R, et al. Harnessing multimodal data integration to advance precision oncology. Nat Rev Cancer 2022;22:114-26. [Crossref] [PubMed]
37. Selvam M, Chandrasekharan A, Sadanandan A, et al. Radiomics as a non-invasive adjunct to Chest CT in distinguishing benign and malignant lung nodules. Sci Rep 2023;13:19062. [Crossref] [PubMed]
38. MacMahon H, Naidich DP, Goo JM, et al. Guidelines for Management of Incidental Pulmonary Nodules Detected on CT Images: From the Fleischner Society 2017. Radiology 2017;284:228-43. [Crossref] [PubMed]
39. Bueno J, Landeras L, Chung JH. Updated Fleischner Society Guidelines for Managing Incidental Pulmonary Nodules: Common Questions and Challenging Scenarios. Radiographics 2018;38:1337-50. [Crossref] [PubMed]
40. Belfiore MP, Sansone M, Monti R, et al. Robustness of Radiomics in Pre-Surgical Computer Tomography of Non-Small-Cell Lung Cancer. J Pers Med 2022;13:83. [Crossref] [PubMed]
41. Rami-Porta R, Bolejack V, Crowley J, et al. The IASLC Lung Cancer Staging Project: Proposals for the Revisions of the T Descriptors in the Forthcoming Eighth Edition of the TNM Classification for Lung Cancer. J Thorac Oncol 2015;10:990-1003.
42. Zhang J, Gold KA, Lin HY, et al. Relationship between tumor size and survival in non-small-cell lung cancer (NSCLC): an analysis of the surveillance, epidemiology, and end results (SEER) registry. J Thorac Oncol 2015;10:682-90. [Crossref] [PubMed]
43. Petty WJ, Urbanic JJ, Ahmed T, et al. Long-Term Outcomes of a Phase 2 Trial of Chemotherapy With Consolidative Radiation Therapy for Oligometastatic Non-Small Cell Lung Cancer. Int J Radiat Oncol Biol Phys 2018;102:527-35. [Crossref] [PubMed]
44. Gomez DR, Tang C, Zhang J, et al. Local Consolidative Therapy Vs. Maintenance Therapy or Observation for Patients With Oligometastatic Non-Small-Cell Lung Cancer: Long-Term Results of a Multi-Institutional, Phase II, Randomized Study. J Clin Oncol 2019;37:1558-65. [Crossref] [PubMed]
45. Sorensen SF, Zhou W, Dolled-Filhart M, et al. PD-L1 Expression and Survival among Patients with Advanced Non-Small Cell Lung Cancer Treated with Chemotherapy. Transl Oncol 2016;9:64-9. [Crossref] [PubMed]
46. Aerts HJ, Velazquez ER, Leijenaar RT, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 2014;5:4006. [Crossref] [PubMed]
47. Yang F, Simpson G, Young L, et al. Impact of contouring variability on oncological PET radiomics features in the lung. Sci Rep 2020;10:369. [Crossref] [PubMed]
48. Kuncman Ł, Pajdziński M, Smółka K, et al. Early lymphocyte levels and low doses radiation exposure of lung predict lymphopenia in radiotherapy for lung cancer. Front Immunol 2024;15:1426635. [Crossref] [PubMed]
49. Windisch P, Guckenberger M. Randomized phase II trial reporting overall survival advantage by adding local consolidative therapy to systemic therapy for oligometastatic non-small cell lung cancer: another step forward on the long road of evidence-based medicine for oligometastatic disease. J Thorac Dis 2019;11:S1869-73. [Crossref] [PubMed]