All procedures of the present study involving human participants were performed in strict compliance with the ethical standards of the institutional research committee and with the 1964 Helsinki declaration and its later amendments or comparable ethical standards. The ethics committee of the First Affiliated Hospital of Zhejiang University (Hangzhou, China) approved the study and waived the requirement for obtaining informed consent.

For all patients included, CT images had been obtained at the First Affiliated Hospital of Zhejiang University (Hangzhou, China). Clinical data were obtained from the medical records of each patient retrospectively. The TNM classification was also extracted and TNM staging was based on the seventh edition of the Union for International Cancer Control and American Joint Committee on Lung Cancer (18).

A 4-µm thick formalin-fixed (at 25°C for 12–24 h), paraffin-embedded (FFPE) tissue slice was used to evaluate ROS1 and ALK gene rearrangement. A total of three 5-µm sections of each tumor were used for total RNA extraction, which was performed based on predefined protocols (19) (RNeasy FFPE kit; Qiagen GmbH). The ROS1 rearrangement status was determined by multiplex real-time PCR in a Stratagene Mx3000P real-time PCR system (Stratagene Corp.) with an AmoyDx^® ROS1 fusion gene detection kit (Amoy Diagnostics Co., Ltd.). ROS1 fusion combinations, including SLC34A2-ROS1, EZR-ROS1, CD74-ROS1, SDC4-ROS1, LRIG3-ROS1, TPM3-ROS1 and GOPC-ROS1, were detected in this assessment. The internal reference gene (β-actin) was used as a negative control and a confirmed ROS1-rearranged DNA was used as a positive control in the AmoyDx^® ROS1 Gene Fusions Detection kit (cat. no. 8.01.23201W012A; Amoy Diagnostics Co., Ltd.).

ALK gene fusion was evaluated with FISH utilizing an ALK break-apart probe (Vysis LSI ALK Dual Color, Break Apart Rearrangement Probe; Abbott Molecular), in accordance with previously published protocols (13).

A PCR-based pyrosequencing assay was used to analyze EGFR mutations located on exons 18–21 (20). The PyroMark system (Qiagen GmbH) was used to perform sequence analysis.

Each patient received thoracic CT scans with either the 64-slice Brilliance, 256-slice iCT (Philips Medical Systems, Inc.) or 64-slice light-speed VCT (GE Healthcare). CT scan systems had the following parameters: Rotation speed, 0.5/sec; pitch, 1; 120 kV; 120–380 mA; field of view, 300–350 mm; and collimation, 0.625 mm. A high-frequency reconstruction algorithm with a 512×512 matrix was used for image reconstruction. The reconstruction thicknesses and intervals were 5.0 and 5.0 mm, respectively. CT imaging information was reviewed in a Philips CT workstation using the mediastinal window setting [width, 200 Hounsfield Units (HU); level, 40 HU] and lung window setting (width, 1,500 HU; level, 500 HU), which were obtained prior to contrast administration.

All CT images were independently reviewed by JL and JX, two radiologists with 10 years of experience. The radiologists were not blinded to the patients' diagnoses of lung adenocarcinoma but were blinded to their genetic statuses. The presence or absence of multiple lesions, lobulated borders, fine spiculated margins, bubble-like lucency, cavity, liquefaction necrosis and air bronchogram were assessed.

Multiple lesions were defined as more than one tumor lesion on a CT scan. A lobulated border was defined as a shallow wavy configuration seen on part of the border of the lesion, excluding lesions that were abutting the pleura (13,20–22). A lesion that possessed fine linear strands extending 1–2 mm radially was defined as having fine spiculated margins. Lesions that possessed air attenuation spots within them were classified as having bubble-like lucency (13,20) and a cavity was defined as a gas-filled space appearing as an area of lower attenuation or lucency (21). A hypodense area on non-enhanced and contrast-enhanced CT images was defined as liquefaction necrosis (23). An air bronchogram was defined as an air-filled (low-attenuation) bronchus against an opaque background (high-attenuation) (13,21).

Lung nodule reconstruction was performed using the 256-slice iCT (Philips Medical Systems), which allowed for the quantification of the volume of individual lesions. The software is able to automatically generate solid tumor boundaries on each cross-section, while extracting intramodular bronchial and cavitating structures in order to reconstruct a three-dimensional structure of the nodule and provide data on the volume (V) and estimated diameter (D) (Fig. 1). For mixed tumor nodules with solid and ground-glass opacity (GGO) portions (Fig. 2A and B), as well as for the entire nodule (Fig. 2C), all computer-derived boundaries were assessed and adjusted as necessary by the two experienced radiologists. The software reconstructed the three-dimensional structure of the nodule and provided data on the entire nodule and the solid portion (Fig. 2D). The radiologists recorded the volume and the estimated diameter mentioned above for the total nodule (tV, tD) and the solid portion (sV, sD). The PSV was calculated using the formula PSV=sV/tV.

The patients with different gene status' were included in different groups via a simple randomization method (Fig. 3). A Chi-squared test was used to compare categorical variables. For continuous variables, including age, tV, tD and PSV, differences in median values between two groups were compared by the Mann-Whitney U-test, while the Kruskal-Wallis U-test was used for comparisons among four groups. The univariate and multivariate logistic regression analyses were performed to predict ROS1/ALK-rearranged. Variates with P≤0.10 in the univariate analysis were included in a multivariable logistic regression analysis. A receiver operating characteristic (ROC) analysis was carried out to evaluate the predictive value of PSV and the multi-factor (all independent predictive factors selected by multivariate analysis) predictive model to predict ROS1 rearrangement or ROS1/ALK rearrangement. All statistical analyses were performed using SPSS version 17.0 (SPSS, Inc.). P<0.05 was considered to indicate statistical significance.

In the present study, 28 ROS1-rearranged, 578 EGFR-mutated and 71 ALK-rearranged patients were retrospectively identified among 1,120 patients with histology-confirmed lung adenocarcinoma diagnosed at the First Affiliated Hospital of Zhejiang University (Hangzhou, China) between March 2008 and October 2015. The 28 ROS1-rearranged patients were assigned to the ROS1 rearrangement group (ROS1 rearrangement group or ROS1⁺ group; n=28). Furthermore, 56 ALK-rearranged patients were randomly selected from the 71 ALK-rearranged patients (ALK rearrangement group or ALK⁺ group; n=56), as well as 112 patients with EGFR mutations from the 578 EGFR-mutated patients (EGFR mutation group or EGFR⁺ group; n=112) and 112 patients from those who exhibited the WT for all three genes (WT group; n=112). The patient selection procedure is presented in Fig. 3. The EGFR mutation group included 65 patients with EGFR 19 exon mutation, 44 patients with EGFR 21 exon mutation, 1 patient with EGFR 20 exon mutation, 1 patient with both EGFR 19 and 20 exon mutation and 1 patient with both EGFR 20 and 21 exon mutation.

The demographic and clinicopathological features of the patients with lung adenocarcinoma are presented in Table I. The median age (range) of the patients in the ROS1⁺, ALK⁺, EGFR⁺ and WT groups was 54.5 (27–60), 51.0 (23–79), 60.0 (34–83) and 62.0 (31–83) years, respectively. Individuals possessing ROS1 rearrangements were noted to be markedly younger in comparison with those with EGFR mutations (P=0.007) and patients who did not test positive for all three genes (P=0.003). Patients with ROS1 mutations were more likely to be female (P=0.006) in contrast to the WT group. Those who had no history of smoking were also more likely to have ROS1 mutations (P=0.005) in comparison with the WT group. The demographic features (age, sex and smoking history) of ROS1-rearranged patients were not different compared to those of ALK-rearranged patients. Patients across the four groups did not exhibit any notable variability in terms of carcinoembryonic antigen levels.

Fig. 4 provides comparisons of the tD, tV and PSV across tumors with various genetic aberrations. The PSV [median (interquartile range)] in tumors with ROS1 rearrangements [87.9 (82.7–92.3) %] was markedly higher compared with those with EGFR mutations [70.4 (51.4–83.4)%; P<0.001] and those of the WT [63.0 (50.9–83.2)%; P<0.001]. However, the difference in PSV between the ROS1⁺ group and the ALK⁺ group [84.0 (70.3–90.0)%; P=0.251] was not significant. There was no difference in the median value of the tD and tV among the four groups.

CT images of four representative patients with different genetic status are presented in Fig. 5. The four tumors exhibited different PSV on CT.

The ROC curves for the predictive value of PSV for the gene rearrangements are provided in Fig. 6. The area under the ROC curve (AUC) of PSV to predict ROS1 rearrangement was 0.736 (95% CI, 0.637–0.835; P<0.001; Fig. 6A) at a cut-off value of 0.849. When the Youden index was maximal, the sensitivity was 0.750 and the specificity was 0.743. Given the similar PSV values in the ROS1⁺ group and the ALK⁺ group, a ROC curve to predict ROS1 or ALK rearrangement was drawn (Fig. 6B). The AUC was 0.702 (95% CI, 0.631–0.773; P<0.001) at a cut-off value of 0.805. When the Youden index was maximal, the sensitivity was 0.697 and the specificity was 0.702.

A total of two radiologists independently reviewed the CT imaging and recorded the interpretations. The differences in interpretation between the two radiologists were as follows: Multiple lesions (3/76; 3.9%), lobulated borders (18/236; 7.6%), fine spiculated margins (19/179; 10.6%), pleural retraction (8/145; 5.5%), bubble-like lucency or cavities (4/42; 9.5%), air bronchograms (7/62; 11.3%) and liquefaction necrosis (1/17; 5.9%). Disagreements were resolved by discussions to reduce the interobserver variability.

Table II summarises the final different features on CT imaging based on the various genetic aberrations. There were no differences in the frequency of multiple lesions, pleural retraction, fine spiculated margins, lobulated borders, bubble-like lucency or cavities, air bronchograms or liquefaction necrosis between tumors in the ROS1⁺ group and tumors in the ALK⁺, EGFR⁺ or WT groups.

Given the similar clinical radiological characteristics between ROS1⁺ tumors and ALK⁺ lncRNAs, and the same first line recommended targeted treatment, univariate and multivariate analyses of clinical radiological parameters were performed to predict ROS1/ALK-rearrangement. All clinical parameters, including age, sex and smoking status, the quantitative volumetric parameters, including the total nodule diameter and PSV and all the aberrations on CT imaging were assessed via univariate logistic regression analysis. The results presented in Table III demonstrated that age, smoking history, total nodule diameter, PSV and existence of lobulated border, fine spiculated margin and pleural retraction were different (P<0.1) between ROS1/ALK-rearranged and non-ROS1/ALK-rearranged patients. These different parameters were subjected to multivariate logistic regression analysis, the results of which are also presented in Table III. Younger age and higher PSV values were independent predictors of ROS1/ALK rearrangements in the multivariate analysis. The correlation analysis between PSV and age suggested that the two factors were not correlated (Fig. S1). The ROC curve of the predictive model combined with age and PSV is presented in Fig. 6C, with an AUC of 0.785.