The value of dual-energy spectral CT in differentiating the pathological grades of T1-size lung adenocarcinoma

Introduction

Background

Lung cancer is the leading cause of death from cancer in the world and lung adenocarcinoma (LUAD) is the most common subtype (1). With the increased implementation of computed tomography (CT) screening with superb spatial resolution, the early detection rate of T1-size lung cancer has risen dramatically (2). The National Comprehensive Cancer Network (NCCN) guidelines recommended radical surgical treatment as the primary treatment for patients with T1-size LUAD, but there has been controversy over the optimal surgical types for patients with T1-size LUAD (3). Among all the factors affecting the choice of surgical type, differentiation grading is a critical pathological index highly related to biological behavior and prognosis (4,5). As a result, compared with T1-size LUAD with a low-grade differentiation, surgery with a relatively larger extent should be performed for high-grade LUAD to reduce recurrence (6,7). Therefore, preoperative accurate grading of T1-size LUAD is of great significance for clinical decision-making.

Rationale and knowledge gap

Currently, histological differentiation grading is based on the percentage of different morphological patterns (8). The guidelines do not require reporting on differentiation grade for the preoperative biopsy in the clinical setting, as needle biopsy specimens are too tiny to represent the entire lesion, which might lead to over-grading or under-grading (9). In addition, due to some possible serious complications such as pneumothorax, hemorrhage, and air embolism, not all patients are suitable for preoperative biopsy (10,11). Thus, there is an urgent need for a non-invasive technique that could accurately predict the differentiation grading for T1-size LUAD before the surgery. Among all the non-invasive techniques, such as CT, magnetic resonance imaging (MRI), and positron emission tomography (PET)-CT, CT is the most commonly used for pathologically grading in the research setting since CT is routinely used in the clinical setting (12-14). In addition, CT-based artificial intelligence (AI), particularly deep learning and radiomics, has revolutionized LUAD identification by enabling high-throughput analysis of multi-modal data, achieving diagnostic accuracy comparable to expert pathologists (15,16). Unlike conventional CT, as a new imaging technology with its unique mechanism, dual energy CT (DECT) can not only provide the morphology characteristics of conventional single-energy CT images but also quantify materials based on their different attenuation properties at various energies, allowing for better identification and characterization of abnormalities or lesions (17-19). Previous studies have found that DECT measurements can help discriminate the pathological grade of lung cancer (20,21). However, their investigation solely utilized contrast-enhanced DECT measurements without incorporating plain DECT measurements. In addition, both studies only included solid lesions without an upper limit on the size.

Objective

Therefore, in this study, we intend to investigate the incremental value of DECT-derived quantitative measurements from plain, arterial phase (AP), and venous phase (VP) DECT in predicting the pathological grades of LUAD in a whole spectrum of density (solid, part-solid, and ground-glass) mimicking a real clinical setting with a focus on only T1-size (≤3 cm) in addition to conventional clinical characteristics.

The manuscript is organized as follows: (I) the methods including patient selection, spectral CT protocol, image analysis, and statistics; (II) results on patient characteristics, feature selection, model performance, and nodule subtype comparisons; (III) clinical implications, compares with prior work, and summarizes limitations for further improvement; (IV) conclusions and future directions. We present this article in accordance with the STARD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-516/rc).

Methods

Patients

This retrospective study was approved by the institutional ethics committee of National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital & Shenzhen Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College (No. KYKT2021-17-1) and individual consent for this retrospective analysis was waived. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments. From October 2018 to January 2022, 137 patients with 161 newly diagnosed LUAD lesions (≤3 cm) having received DECT were retrospectively enrolled. The inclusion criteria were as follows: (I) age ≥18 years old; (II) histologically confirmed LUAD after surgery; (III) the maximum diameter of the lesion ≤3 cm; (IV) no anti-tumor treatment received before CT scan; (V) both AP and VP DECT available. The exclusion criteria were as follows: (I) history of other malignancies; (II) patients with incomplete clinical and pathological data; (III) patients with images of low quality. The histological grades were assessed by a pathologist with 10 years of experience according to tumor histological characteristics, and the resected tumors were classified into well-differentiated (n=41), moderately differentiated (n=104), and poorly differentiated (n=16) groups based on the 2020 International Association for the Study of Lung Cancer (IASLC) novel grading system (8). Because there were a limited number of patients with poorly differentiated lesions, the enrolled patients were further reclassified into a low-grade group (including a well-differentiated group) and a high-grade group (including poorly differentiated and moderately differentiated groups).

CT scanning protocol

All examinations were performed on a 256-row multidetector CT scanner (Revolution CT, GE HealthCare, Milwaukee, WI, USA) using gemstone spectral imaging (GSI) mode. The CT scan parameters were as follows: scanning field, 50 cm; slice thickness, 5 mm; slice spacing, 5 mm; matrix =512×512; voltages of 140 and 80 kVp with ultra-fast peak kVp switching (0.5 ms) at automated tube current; helical pitch =0.992; collimation =64 × 0.625 mm; rotation time =0.5 ms; kernel = standard. The scanning range was from the lung tip to the lung bottom. According to the patient’s body weight, a bolus of the contrast medium (Loversol, 350 mg/mL) was injected at a standard dosage of 1.5 mL/kg and a flow rate of 2.0–3.0 mL/s. AP and VP scans were performed at 30 and 60 s after contrast injection using automatic tracking technology. The CT images were reconstructed using an adaptive statistical iterative reconstruction-Veo (ASIR-V) at 50% weight algorithm with a reconstructed slice thickness of 1.25 mm.

CT image analysis

Raw CT images were transferred to an Advanced Workstation (AW4.7, GE HealthCare). A radiologist with over 10 years of experience in thoracic imaging, who was blind to the pathological results initially, assessed the CT images. Another senior radiologist who was also blind to the pathological results reviewed the CT images, with any discrepancy resolved through discussion. The following information was obtained including the location of the tumor (right upper lobe, right lower lobe, right middle lobe, left upper lobe, and left lower lobe), tumor density (solid, part-solid, and pure ground-glass), shape (regular and irregular), presence of vascular convergence sign, spiculation sign, air bronchogram sign, pleural indentation sign, and lobulation sign.

Quantitative measurements

Two experienced radiologists performed the quantitative measurements. For the part-solid and solid nodules, a round region of interest (ROI) was manually placed in the solid components of the lesion, avoiding areas of calcification, necrosis, and blood vessels that could affect the measurement results. For pure ground-glass nodules, a round ROI was placed in the ground-glass component, avoiding visible blood vessels. To ensure consistency, the axial slice with the largest area of the lesion and its adjacent slices (above and below) were selected for measurement, and then the three values were averaged. Finally, the two radiologists’ measurements were averaged again to obtain the final measurement. The measurements included iodine concentration (IC), effective atomic number (Z_eff), and CT values of 40–140 keV. The slope of the spectral curve of the lesion was defined as λ_HU = (CT_40keV − CT_100keV)/60. To minimize the influence of the individual blood circulation status and scanning timing, the IC values of lung lesions were normalized to that of the aorta in the same section to calculate the normalized iodine concentration (NIC): NIC = IC_lesion (IC_l)/IC_aorta (IC_a). The IC difference (ICD) between the contrast-enhanced lesion and the unenhanced lesion was also calculated using the formula: ICD = IC_enhanced − IC_unenhanced.

Statistical analysis

SPSS version 26.0 (IBM, Armonk, NY, USA) was used for statistical analysis. Normally distributed data were reported as the mean ± standard deviation (SD), and non-normally distributed data were reported as median and interquartile range (IQR). The t-test or Mann-Whitney U test was used for comparing continuous data between two groups. To compare three or more groups, one-way analysis of variance and Kruskal-Wallis test were used. Chi-squared test was used to test the difference for categorical variables. Intraclass correlation coefficients (ICCs) were used to assess the reproducibility of the measurements by the two radiologists, and any measurements with ICC ≥0.8 were considered reliable. Pearson’s correlation was used to test the correlation between spectral features within each of the plain, AP, and VP groups, respectively. To address multicollinearity issues, we employed a stepwise regression algorithm to independently screen significant variables derived from plain, AP, and VP spectral CT. Subsequently, we integrated the selected features from all three phases to construct the spectral CT model. A clinical model was constructed using variables with significant univariate analysis results as independent variables and pathological grading as the dependent variable. A combined model was developed by integrating spectral CT features with clinical features as independent variables, while maintaining pathological grading as the dependent variable. Receiver operating characteristic (ROC) analysis was used to evaluate these models’ performance by calculating the area under the curve (AUC), sensitivity, and specificity. These prediction models’ optimal sensitivity and specificity were calculated based on maximizing the Youden index (sensitivity + specificity − 1). The DeLong test (22) was used to compare the performance between each of these three models. P<0.05 was considered statistically significant.

Results

Patient characteristics

In this study, 137 newly diagnosed LUAD patients with 161 lesions (41 low-grade lesions and 120 high-grade lesions) were included. There were 51 males and 86 females, with an average age of 55.55±12.71 years. The median tumor long-axis diameter was 1.40 (IQR, 0.85, 1.90) cm and the tumor short-axis diameter was 1.10 (IQR, 0.60, 1.45) cm. The flowchart of patient selection is shown in Figure 1, and the clinical characteristics of enrolled patients are summarized in Table 1. There were significant differences in tumor density (P<0.001), tumor long-axis diameter (P<0.001), tumor short-axis diameter (P<0.001), shape (P=0.009), presence of spiculation sign (P=0.01), and lobulation sign (P=0.01) between the low-grade LUAD and the high-grade LUAD. Multivariate logistic regression analysis revealed significant associations between pathological grading and tumor density patterns including part-solid [odds ratio (OR) =0.077; 95% confidence interval (CI): 0.010–0.620; P=0.02] and pure ground glass (OR =0.061; 95% CI: 0.007–0.524; P=0.01) compared to the reference category of solid nodules (Table 1).

Figure 1 The flowchart of patient selection. CT, computed tomography.

Feature selection and correlation

All the ICCs of measured results from the two radiologists were greater than 0.8, which indicates excellent agreement (Table S1). The CT values of 40–140 keV and Z_eff of the high-grade group were significantly higher than those of the low-grade group in plain, AP, and VP (all P<0.001) (Figure 2 and Tables 2-4). In addition, in the plain, AP, and VP CT, the mono-energy CT values of both the low-grade and the high-grade groups decreased along with increasing energy levels, but the rate of decline gradually slowed down (Figure 3). As shown in Figure 4, all features extracted from plain CT were highly correlated with each other with correlation coefficients ranging from 0.891 to 1.000 in plain DECT (Figure 4A). The same is true for AP spectral CT (0.891 to 1.000) (Figure 4B) and VP DECT (0.870 to 1.000) (Figure 4C). Stepwise regression analysis was performed separately for features extracted from plain DECT, AP DECT, and VP DECT, and results showed that CT_40keV in plain DECT, CT_140keV in AP DECT, and CT_140keV in VP DECT were independently significant predictors for the pathological grade of T1-size LUAD. Separate stepwise regression analyses were conducted for CT values (40–140 keV) and Z_eff measurements obtained during plain DECT, AP DECT, and VP DECT, and results showed that CT_40keV in plain DECT (β=0.394, P<0.001), CT_140keV in AP DECT (β=0.399, P<0.001), and CT_140keV in VP DECT (β=0.408, P<0.001) were independently significant predictors for the pathological grade of T1-size LUAD.

Figure 2 Comparison CT value at 40–140 keV derived from plain, AP, and VP spectral CT. (A) Plain spectral CT, (B) AP spectral CT, (C) VP spectral CT. AP, arterial phase; CT, computed tomography; VP, venous phase.

Figure 3 Comparative CT imaging and spectral curve analysis of representative cases from the low-grade and high-grade groups. (A) CT images and spectral curves of a 51-year-old male with poorly differentiated (a solid nodule). (A1) The non-enhanced image in lung window settings. (A2) The non-enhanced image in mediastinal window settings. (A3) The spectral curve in plain, AP, and VP. (B) CT images and spectral curves of a 47-year-old female with well-differentiated (a part-solid nodule). (B1) The non-enhanced image in lung window settings. (B2) The non-enhanced image in mediastinal window settings. (B3) The spectral curve in plain, AP, and VP. AP, arterial phase; CT, computed tomography; HU, Hounsfield unit; VP, venous phase.

Figure 4 Correlation heatmaps regarding CT value at 40–140 keV and Z_eff during the plain, AP, and VP spectral CT. (A) Plain spectral CT, (B) AP spectral CT, (C) VP spectral CT. AP, arterial phase; CT, computed tomography; VP, venous phase.

Model performance

The clinical prediction model incorporated six predictors: tumor density, tumor long diameter, tumor short diameter, shape, spiculation sign, and lobation sign. The clinical model achieved good performance with an AUC of 0.772 (95% CI: 0.699–0.844), an optimal sensitivity of 0.567, and a specificity of 0.902. The spectral CT-based predictive model incorporated three dependent parameters: CT_40keV in plain, CT_140keV in AP, and CT_140keV in VP. The diagnostic performance of the spectral CT model was slightly improved (AUC =0.774, 95% CI: 0.700–0.849, sensitivity =0.658, and specificity =0.805), respectively. When clinical variables were combined with all the selected measurements from the DECT, the combined model achieved the best performance (AUC =0.825, 95% CI: 0.759–0.890, sensitivity =0.633, specificity =0.927), which significantly outperformed the clinical model (P=0.03), but insignificantly outperformed the spectral model (P=0.07). The ROC curve results are presented in Table 5 and Figure 5.

Figure 5 Receiver operating characteristic analysis for the clinical model, spectral CT model, and combined model. CT, computed tomography.

Comparison of DECT parameters among solid nodules, part-solid nodules, and ground-glass nodules

In this study, there were 42 solid nodules, 78 part-solid nodules, and 41 ground-glass nodules. During the plain scan, there were no statistically significant differences in IC_l (P=0.12) and λ_HU (P=0.03) among the three groups. During the AP scan, there was no statistically significant difference in IC_l (P=0.06), ICD (P=0.58), NIC (P=0.10), and λ_HU (P=0.06) among the three groups (Table S2). During the VP scan, the IC_l, ICD, NIC, and λ_HU of the solid nodules were significantly higher than those of the pure ground-glass nodules (all P<0.001). These parameters of part-solid nodules were also significantly greater than those of pure ground-glass nodules (IC_l, P<0.001; ICD, P=0.002; NIC, P<0.001; λ_HU, P<0.001). However, there was no statistically significant difference in IC_l (P=0.85), ICD (P=0.40), NIC (P=0.21), and λ_HU (P=0.80) between solid nodules and part-solid nodules (Figure 6). In addition, the CT values of 40–140 keV and Z_eff of the solid nodules group and part-solid nodules group were significantly higher than those of the ground-glass nodules group in plain, AP, and VP DECT. These measurements of solid nodules are also significantly higher than those of part-solid nodules (all P<0.001) (Table S3).

Figure 6 Comparison of spectral CT parameters of solid nodules, part-solid nodules, and ground-glass nodules in plain scan, arterial, and venous phases. (A) IC_l; (B) λ_HU; (C) NIC; (D) ICD. *, the difference between the pairwise comparison results is statistically significant. CT, computed tomography; IC_l, iodine concentration of lesion; ICD, iodine concentration difference; NIC, normalized iodine concentration; λ_HU, the slope of spectral curve.

Discussion

Key findings, comparison with similar researches, and explanations of findings

In this study, we investigated the value of clinical features and DECT measurements in differentiating the pathological grades of T1-size LUAD. We found that lesion density and single-energy CT values are independent predictors for differentiating the pathological grade of T1-size LUAD. The clinical model, spectral CT model, and combined model achieved good performance, and combining clinical features and features extracted from plain, AP, and VP spectral CT gained a significant incremental benefit.

In this study, the CT values of 40–140 keV and Z_eff of the low-grade group were significantly lower than those of the high-grade group, which is inconsistent with a previous study by Mu et al. (20). Unlike their study, which only included solid nodules, we enrolled a large percentage of part-solid and ground-glass nodules, especially in the low-grade group (40/41), resulting in a considerably lower mean Z_eff value in the low-grade group in comparison to their mean Z_eff value in the well-differentiated group. In theory, CT value and Z_eff are positively correlated with the density of matter, which was supported by a previous study (23) and our result that solid nodules had the highest CT value and Z_eff value, followed by part-solid and ground-glass nodules. This could further support our findings that the high-grade LUAD group with a considerably larger percentage of solid nodules had significantly higher mean CT value and Z_eff value than that of the low-grade LUAD group. Simultaneously, multivariate logistic regression analysis showed that the nodule density type was an independent predictor for pathological grades. Therefore, nodule density is the most important factor for predicting the pathological grade of T1-size LUAD, and DECT provides another way to characterize the pathological grade more quantitatively. Also, we found that the CT values at different single-energy levels and Z_eff values were highly correlated with each other within each of the plain, AP, and VP feature sets. As a result, stepwise regression analysis showed that one CT value is enough to represent other CT values and Z_eff value for the differentiation. The CT values at 40 keV in plain scan, 140 keV in AP, and 140 keV in VP demonstrate significant discriminative value in differentiating between low-grade and high-grade LUAD. This can be explained through the following mechanisms. First, the low-energy 40 keV imaging in plain scan significantly enhances soft tissue contrast resolution (24). This improved contrast allows for better differentiation of the fundamental differences between low-grade (predominantly lepidic growth) and high-grade (solid pattern and micropapillary pattern) LUAD at their baseline state. Second, the high-energy 140 keV imaging during AP and VP reduces the energy spectrum-dependent attenuation of iodine contrast agents, enabling more accurate enhancement evaluation (25). The high monochromaticity of 140 keV minimizes beam-hardening artifacts, resulting in more reliable quantitative analysis of enhancement patterns. When combining the three features from plain, AP, and VP, the AUC value showed further improvement, comprehensively reflecting both the spectral attenuation characteristics and hemodynamic behavior of LUAD.

In this study, there was no statistical difference in λ_HU in the plain, AP, and VP CT between the two groups, respectively. However, previous research shows that the λ_HU of the low-grade group was significantly higher than that of the high-grade group in AP and VP (20,21). The possible reasons for this inconsistency are due to the difference in the included lesion histology, inclusion criteria, lesion density, lesion size, ROI selection methods, grouping criteria, etc. Among them, the most significant difference is that we enrolled nodules with different densities in our study. In this study, there was only 1 case of solid nodule in the low-grade group and 41 cases in the high-grade group. Thus, it is not possible to do a subgroup analysis on the difference in λ_HU between the solid nodules of the low-grade group and the solid nodules of the high-grade group like previous studies. In addition, we found that in the VP CT, the λ_HU of solid nodules and part-solid nodules was significantly higher than that of ground-glass nodules, suggesting the inappropriateness of directly comparing our results with previous studies. Therefore, this consistency with previous studies warrants further study with a larger sample size and nodules in different densities to further confirm our findings.

In different phases, changes in IC can reflect the hemodynamic changes of the lesion by tracking the change in IC. Evidence has shown that the IC generated from spectral CT had excellent agreement with the actual contrast agent concentration (26). In our study, there were no statistically significant differences in IC, ICD, and NIC between the low-grade group and the high-grade group in the plain, AP, and VP DECT. This is inconsistent with previous research results (20,21), that the IC and NID of the high-grade group were significantly lower than those of the low-grade group in AP and VP DECT. This inconsistency between our study and the previous two studies might be largely due to nodule size and nodule density. We focused on T1-size LUAD presented as solid, part-solid, and ground-glass nodules. In contrast, both previous studies enrolled LUAD (>2 cm) only presented as solid nodules. IC of malignant solid lung nodules was shown to be negatively correlated with the diameter, speculating necrosis is more frequently present in larger solid lesions leading to lower IC (27). However, in our study, necrosis is rarely present in T1-size LUAD. That is probably why the average IC and NIC of the solid nodules in our study were greater than the average IC and NIC in the study by Mu et al. In addition, in terms of nodule density, in our study, there were many solid nodules and part-solid nodules both in the high-grade group (96/120) and the low-grade group (24/41). And the IC_l, ICD, and NIC of the solid nodules and part-solid nodules were significantly higher than those of the ground-glass nodules in VP CT, but not in the plain and AP CT. Consistently, Liu et al. also found that there was no significant difference in IC value between pure ground-glass nodules and part-solid ground-glass nodules during AP, however, they did not investigate the difference in VP spectral CT (28). Therefore, it is also not appropriate to directly compare our results with the two previous studies, and our results need further study with a larger sample size to confirm.

Limitations

Our study has several limitations. First, the retrospective design, small sample size, and single-center setting may contribute to selection bias and limit the generalizability of our results. Future multicenter prospective studies with larger samples are warranted to validate these observations. Second, although three consecutive slices of ROI were analyzed, the limited sections may not represent the entire tumor. Although we set the rules to measure the solid components for all part-solid nodules, this measurement method might be biased due to the loss of information on ground-glass components.

Future research directions

These quantitative spectral CT parameters may have broader implications in the era of AI-assisted diagnosis. Recent advances in AI have demonstrated significant potential to enhance LUAD detection and characterization through two primary mechanisms: (I) deep learning architectures, particularly convolutional neural networks (CNNs), have achieved notable success in automated feature extraction from CT imaging data. The spectral CT biomarkers identified in our study could serve as standardized, quantitative inputs to further optimize these computational models (29). (II) AI frameworks now enable sophisticated integration of radiological features with histopathological and genomic data (30). Our spectral parameters may provide complementary value to existing radiogenomic models. The integration of spectral CT parameters with AI methods could significantly improve the accuracy and clinical utility of LUAD grading systems in future studies.

Conclusions

Single-energy CT values have the potential to predict the pathological grades of T1-size LUAD in the clinical real world. Both clinical features and quantitative spectral measurements are valuable in differential diagnosis. Future studies could integrate the quantitative parameters of spectral CT with deep learning architectures to develop combined analytical models.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the STARD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-516/rc

Data Sharing Statement: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-516/dss

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-516/prf

Funding: This work was supported by Shenzhen Clinical Research Center for Cancer (No. [2021] 287); Shenzhen Science and Technology Program (grant/award number: KCXFZ20201221173008022); National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital & Shenzhen Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College, Shenzhen (No. SZ2020QN001); and Shenzhen High-level Hospital Construction Fund.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-516/coif). K.M. is currently an employee of GE HealthCare and has provided technical support in this study. 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. This retrospective study was approved by the institutional ethics committee of National Cancer Center/National Clinical Research Center for Cancer/Cancer Hospital & Shenzhen Hospital, Chinese Academy of Medical Sciences and Peking Union Medical College (No. KYKT2021-17-1) and individual consent for this retrospective analysis was waived. This study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

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. Siegel RL, Miller KD, Wagle NS, et al. Cancer statistics, 2023. CA Cancer J Clin 2023;73:17-48. [Crossref] [PubMed]
2. Henschke CI, Yip R, Shaham D, et al. A 20-year Follow-up of the International Early Lung Cancer Action Program (I-ELCAP). Radiology 2023;309:e231988. [Crossref] [PubMed]
3. Ettinger DS, Wood DE, Aisner DL, et al. Non-Small Cell Lung Cancer, Version 3.2022, NCCN Clinical Practice Guidelines in Oncology. J Natl Compr Canc Netw 2022;20:497-530. [Crossref] [PubMed]
4. Fujikawa R, Muraoka Y, Kashima J, et al. Clinicopathologic and Genotypic Features of Lung Adenocarcinoma Characterized by the International Association for the Study of Lung Cancer Grading System. J Thorac Oncol 2022;17:700-7. [Crossref] [PubMed]
5. Li Y, Liu J, Yang X, et al. An ordinal radiomic model to predict the differentiation grade of invasive non-mucinous pulmonary adenocarcinoma based on low-dose computed tomography in lung cancer screening. Eur Radiol 2023;33:3072-82. [Crossref] [PubMed]
6. Saji H, Okada M, Tsuboi M, et al. Segmentectomy versus lobectomy in small-sized peripheral non-small-cell lung cancer (JCOG0802/WJOG4607L): a multicentre, open-label, phase 3, randomised, controlled, non-inferiority trial. Lancet 2022;399:1607-17. [Crossref] [PubMed]
7. Suzuki K, Watanabe SI, Wakabayashi M, et al. A single-arm study of sublobar resection for ground-glass opacity dominant peripheral lung cancer. J Thorac Cardiovasc Surg 2022;163:289-301.e2. [Crossref] [PubMed]
8. Moreira AL, Ocampo PSS, Xia Y, et al. A Grading System for Invasive Pulmonary Adenocarcinoma: A Proposal From the International Association for the Study of Lung Cancer Pathology Committee. J Thorac Oncol 2020;15:1599-610. [Crossref] [PubMed]
9. Yang X, Dong X, Wang J, et al. Computed Tomography-Based Radiomics Signature: A Potential Indicator of Epidermal Growth Factor Receptor Mutation in Pulmonary Adenocarcinoma Appearing as a Subsolid Nodule. Oncologist 2019;24:e1156-64. [Crossref] [PubMed]
10. Yucel S, Sayit AT, Tomak L, et al. Frequency of complications and risk factors associated with computed tomography guided core needle lung biopsies. Ann Saudi Med 2021;41:78-85. [Crossref] [PubMed]
11. Lee JH, Yoon SH, Hong H, et al. Incidence, risk factors, and prognostic indicators of symptomatic air embolism after percutaneous transthoracic lung biopsy: a systematic review and pooled analysis. Eur Radiol 2021;31:2022-33. [Crossref] [PubMed]
12. Jiang J, Xiao Y, Liu J, et al. T1 mapping-based radiomics in the identification of histological types of lung cancer: a reproducibility and feasibility study. BMC Med Imaging 2024;24:308. [Crossref] [PubMed]
13. Shen H, Chen L, Liu K, et al. A subregion-based positron emission tomography/computed tomography (PET/CT) radiomics model for the classification of non-small cell lung cancer histopathological subtypes. Quant Imaging Med Surg 2021;11:2918-32. [Crossref] [PubMed]
14. Cheng DO, Khaw CR, McCabe J, et al. Predicting histopathological features of aggressiveness in lung cancer using CT radiomics: a systematic review. Clin Radiol 2024;79:681-9. [Crossref] [PubMed]
15. Lu MY, Chen TY, Williamson DFK, et al. AI-based pathology predicts origins for cancers of unknown primary. Nature 2021;594:106-10. [Crossref] [PubMed]
16. Yang R, Li W, Yu S, et al. Deep Learning Model for Pathological Grading and Prognostic Assessment of Lung Cancer Using CT Imaging: A Study on NLST and External Validation Cohorts. Acad Radiol 2025;32:533-42. [Crossref] [PubMed]
17. Tatsugami F, Higaki T, Nakamura Y, et al. Dual-energy CT: minimal essentials for radiologists. Jpn J Radiol 2022;40:547-59. [Crossref] [PubMed]
18. Lennartz S, Hokamp NG, Kambadakone A. Dual-Energy CT of the Abdomen: Radiology In Training. Radiology 2022;305:19-27. [Crossref] [PubMed]
19. Albano D, Di Luca F, D'Angelo T, et al. Dual-energy CT in musculoskeletal imaging: technical considerations and clinical applications. Radiol Med 2024;129:1038-47. [Crossref] [PubMed]
20. Mu R, Meng Z, Guo Z, et al. Dual-layer spectral detector computed tomography parameters can improve diagnostic efficiency of lung adenocarcinoma grading. Quant Imaging Med Surg 2022;12:4601-11. [Crossref] [PubMed]
21. Lin LY, Zhang Y, Suo ST, et al. Correlation between dual-energy spectral CT imaging parameters and pathological grades of non-small cell lung cancer. Clin Radiol 2018;73:412.e1-7. [Crossref] [PubMed]
22. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach. Biometrics 1988;44:837-45.
23. Ghetti C, Ortenzia O, Bertolini M, et al. Lung dual energy CT: Impact of different technological solutions on quantitative analysis. Eur J Radiol 2023;163:110812. [Crossref] [PubMed]
24. Greffier J, Villani N, Defez D, et al. Spectral CT imaging: Technical principles of dual-energy CT and multi-energy photon-counting CT. Diagn Interv Imaging 2023;104:167-77. [Crossref] [PubMed]
25. Hamid S, Nasir MU, So A, et al. Clinical Applications of Dual-Energy CT. Korean J Radiol 2021;22:970-82. [Crossref] [PubMed]
26. Harsaker V, Jensen K, Andersen HK, et al. Quantitative benchmarking of iodine imaging for two CT spectral imaging technologies: a phantom study. Eur Radiol Exp 2021;5:24. [Crossref] [PubMed]
27. Aoki M, Takai Y, Narita Y, et al. Correlation between tumor size and blood volume in lung tumors: a prospective study on dual-energy gemstone spectral CT imaging. J Radiat Res 2014;55:917-23. [Crossref] [PubMed]
28. Liu G, Li M, Li G, et al. Assessing the Blood Supply Status of the Focal Ground-Glass Opacity in Lungs Using Spectral Computed Tomography. Korean J Radiol 2018;19:130-8. [Crossref] [PubMed]
29. Quanyang W, Yao H, Sicong W, et al. Artificial intelligence in lung cancer screening: Detection, classification, prediction, and prognosis. Cancer Med 2024;13:e7140. [Crossref] [PubMed]
30. Shao X, Ge X, Gao J, et al. Transfer learning-based PET/CT three-dimensional convolutional neural network fusion of image and clinical information for prediction of EGFR mutation in lung adenocarcinoma. BMC Med Imaging 2024;24:54. [Crossref] [PubMed]