Preoperative CT semantic features predict lymph node metastasis of part-solid nodules in non-small cell lung cancer: a multicenter retrospective study

doi:10.21203/rs.3.rs-2743272/v1

Download PDF

Research Article

Preoperative CT semantic features predict lymph node metastasis of part-solid nodules in non-small cell lung cancer: a multicenter retrospective study

https://doi.org/10.21203/rs.3.rs-2743272/v1

This work is licensed under a CC BY 4.0 License

Version 1

posted

You are reading this latest preprint version

Background

Lymph node metastasis (LNM) is the most common way of metastasis of lung cancer, and it is an independent risk factor for long-term survival and recurrence of non-small cell lung cancer (NSCLC) patients. The purpose of this study was to explore the value of preoperative computed tomography (CT) semantic features in differential diagnosis of LNM in part-solid nodules of NSCLC.

Methods

A total of 955 NSCLC patients confirmed by postoperative pathology were retrospectively enrolled from January 2019 to March 2023. The clinical, pathological data and preoperative CT images of these patients were investigated and statistically analyzed in order to explore the risk factors of LNM. Multivariate logistic regression was used to select independent risk factors and establish different prediction models. 10-fold cross-validation was used for model training and validation. The area under the curve (AUC) of receiver operating characteristic curve (ROC) was calculated and the Delong test was performed to compare the predictive performance between models.

Results

LNM occurred in 68 of 955 patients. After univariate analysis and adjustment for confounding factors, smoking history, pulmonary disease, solid component proportion, pleural contact type, and mean diameter were screened as independent risk factors for differential LNM. The image predictors model established by four independent factors of CT semantic features except smoking history showed a good diagnostic efficiency for LNM. Its AUC in the validation group was 0.857, and the sensitivity, specificity and accuracy of the model were all 77.6%.

Conclusions

Preoperative CT semantic features have good diagnostic value for LNM of NSCLC. The image predictors model based on pulmonary disease, solid component proportion, pleural contact type and mean diameter has excellent diagnostic efficacy, and can provide non-invasive evaluation for clinical practice.

Lymph node metastasis

Non-small cell lung cancer

Part-solid nodule

Computed tomography

Predictive model

Lung cancer is still the malignant tumor with the highest morbidity and mortality in the world, of which non-small cell lung cancer (NSCLC) accounts for about 85% of the total patients [1–4]. Lymph node metastasis (LNM) is a common mode of metastasis in lung cancer, and it is an independent risk factor for long-term survival and recurrence of NSCLC patients with 5-year survival rates of 49% for N1 disease, 36% for N2 disease, and 20% for N3 disease [5–7]. Preoperative prediction of LNM will directly affect the method of intraoperative lymph node dissection. The methods of lymph node resection include systematic lymph node dissection, specific lymph node dissection, systematic lymph node sampling, lymph node sampling and so on [8, 9]. Appropriate method of lymph node dissection can maximize the removal of tumor cells on the basis of accurate tumor-node-metastasis (TNM) staging and avoid excessive dissection.

With the popularization of computed tomography (CT) screening technology, more and more patients with early lung cancer are found and treated by surgery [10, 11]. Pulmonary nodules are classified as solid, pure ground-glass, and part-solid nodules (PSNs) based on CT phenotyping [12]. PSN was defined as a nodule containing both ground-glass opacity (GGO) and a solid portion [13]. The Fleischner Society pointed out in its 2017 report that from a therapeutic point of view, pure and part-solid ground glass nodules are best considered as a category separate from purely solid lesions [14].

Pathological examination is the gold standard for the diagnosis of lymph node metastasis in NSCLC, but its reproducibility is quite different and invasive operation may cause negative effects on patients such as tumor cells spread [15–17]. The metastasis rate of lymph nodes in PSNs has been known to be low, which makes the intraoperative evaluation of lymph nodes a controversial issue [18]. Therefore, a non-invasive method to accurately judge the status of lymph nodes before operation is necessary and urgently needed. As a non-invasive and reproducible imaging examination, CT is the most commonly used means to evaluate lymph nodes and is most widely used in clinical practice. The purpose of this study was to explore the risk factors of LNM in preoperative CT semantic features of NSCLC patients manifested as PSNs and to establish a predictive model with application value.

2.1. Patients

This retrospective study was approved by the Ethics Committee of Tongde Hospital of Zhejiang Province (No.2022-029-JY) and the Ethical Committee of the First Affiliated Hospital of Bengbu Medical College (No.2021 − 292), and waived the need of informed consent. The study was conducted in accordance with the Declaration of Helsinki.

Using hospital electronic medical record system, we retrieved the files of all consecutive patients with a pathologic diagnosis of primary NSCLC surgically resected from January 2019 to March 2023. A total of 955 patients were enrolled according to the following inclusion criteria: (1) Radical surgery and lymph node dissection were performed, and NSCLC was confirmed by postoperative pathology; (2) patients underwent unenhanced chest CT scan within two weeks before operation; (3) CT images were satisfactory for analysis; (4) the lesion appeared as part-solid nodule. The exclusion criteria were as follows: (1) patients received radiotherapy or chemotherapy before surgery; (2) CT scan thickness more than 2 mm; (3) clinical and pathological data could not be found or incomplete; (4) patients had a history of malignant tumor. According to the presence or absence of lymph node metastasis, the patients were divided into LNM group and N-LNM group (Fig. 1). Clinical information including age, gender, smoking history, surgery history and clinical basic disease was obtained by searching medical records.

2.2. CT image acquisition

Non-enhanced chest CT scan was performed by seven CT scanners (SOMATOM Definition Flash, FORCE CT, Sensation 16, Definition AS 40, Siemens Medical Solutions, Forchheim, Germany; Revolution, Optima 680, LightSpeed VCT XT, GE Medical Systems, Waukesha, Wis), and all images were reconstructed with a slice thickness of 1–2 mm. All patients were scanned in supine position, arms up, head advanced and breath-holding at the end of deep inhale, scanned from thoracic entrance to adrenal level. The scanning parameters are as follows: tube voltage of 120 kV, automatic tube current modulation technology, collimation of 0.625 mm, pitch of 0.984, helical scanning speed of 0.6 ~ 0.8 s/r, image matrix of 512⋅512, field of view (FOV) under 240 mm⋅240 mm. The width of the lung window was 1500 HU and the window level was − 400 HU, and the width of the mediastinal window was 400 HU and the window level was 40 HU. After scanning, the data were transmitted to the workstation of PACS system.

2.3. Imaging evaluation

Blinded to the clinic-pathologic data, two senior thoracic imaging diagnostic physicians (with 17 and 12 years of experience respectively) independently interpreted the following CT features: the location, shape, and solid component proportion of the nodules; lobulation; spiculation; air cavity; pulmonary disease; air bronchogram type and pleural contact type. Lobulation shows as the irregular undulation of the nodule margin [19]. Spiculation was defined as linear strands that extend from the nodule surface into the lung parenchyma without reaching a pleural surface [20], and it was caused by the contraction and traction of the surrounding interlobular septum when the tumor cells invade and extend outward. The types of air cavity included vacuole, cavity, and cavum. The air cavity with diameter less than 5 mm was defined as vacuole, otherwise it was a cavity; cavum referred to the thick-walled air cavity. Pulmonary disease included emphysema, bullae, and ventilation-perfusion imbalance. Air bronchogram types were classified into bronchial non-deformation, bronchial deformation, and adjacent bronchiectasis. Pleural contact types contained direct contact and pleural traction by tags with indentation or not. Pleural tag was defined as a line or lines extending from a nodule to the pleural surface due to thickening of the interlobular septum [21]. Besides, mean diameter was measured with the average of the long and short diameters; on the premise of avoiding the areas of blood vessels, calcification and cystic degeneration, we delineated the region of interest (ROI) on the highest visible attenuation of lesions and measured the mean of CT attenuation as CT_mean value. Any disagreement in describing semantic features was resolved by a consensus read (Fig. 2).

2.4. Statistical analysis

All statistical analyses were performed using the SPSS software (Version 25.0.0, 64-bit) and R software (Version 4.2.2, https://www.r-project.org). The normality of data was determined by Kolmogorov–Smirnov test. In the univariate difference analysis, the quantitative data conforming to the normal distribution were expressed as means \(\pm\) standard deviations and were compared using the independent sample t-test; the quantitative data not conforming to the normal distribution were expressed as medians with their interquartile range and were compared using the Mann-Whitney U test; Chi-square test and Fisher's exact probability test were used to compare the differences of qualitative data. Multivariate logistic regression was used to adjust the variables with differences in univariate difference analysis to eliminate confounding factors, then independent risk factors were screened out and different prediction models were established. 10-fold cross-validation was used for model training and validation. The efficiency of each model was evaluated by the receiver operating characteristic (ROC) curves, and the area under the curve (AUC) was also calculated. The Delong test was used to compare the differences between the models. A P value of less than 0.05 was considered to indicate statistical significance.

3.1. Clinical characteristics of patients in training group

Among the 799 cases patients in training group, 58 cases (7.3%) occurred LNM (mean age, 60.9\(\pm\)9.4 years; range, 41–91 years; 34 females and 24 males), including 56 cases of adenocarcinoma and 2 cases of squamous carcinoma; 741 cases (92.7%) manifested as N-LNM (mean age, 60.7\(\pm\)9.7 years; range, 18–86 years; 476 females and 265 males), including 738 cases of adenocarcinoma and 3 cases of adenosquamous carcinoma. A total of 9933 lymph nodes were detected in postoperative samples from 799 patients, involving 9716 (97.8%) negative lymph nodes and 217 (2.2%) cancerous lymph nodes (Fig. 3).

In univariate difference analysis, there was significant difference in the distribution of smoke status between the two groups (P = 0.021), and it was shown to be an independent risk factor for LNM after multivariate logistic regression (P = 0.049). There was no significant difference in the distribution of gender, age, surgical history and clinical basic disease between the LNM group and N-LNM group (P＞0.05). The analytical results of clinical data for all patients are summarized in Table 1.

Table 1

Baseline clinical characteristics of patients.
Variable	Univariate						Multivariate
	Training Group			Validation Group			Multivariate
	N-LNM	LNM	P value	N-LNM	LNM	P value	P value
Total cases	741(92.7)	58(7.3)		146(93.6)	10(6.4)
Gender			.391			.458	NA
Female	476(64.2)	34(58.6)		92(63.0)	8(80.0)
Male	265(35.8)	24(41.4)		54(37.0)	2(20.0)
Age^*	60.7 ± 9.7	60.9 ± 9.4	.878	61.1 ± 10.7	60.9 ± 9.2	.964	NA
Smoke			.021			1.000	.049
No	662(89.3)	46(79.3)		127(87.0)	9(90.0)
Yes	79(10.7)	12(20.7)		19(13.0)	1(10.0)
Surgery history			.109			.348	NA
No	542(73.1)	48(82.8)		74(50.7)	3(30.0)
Yes	199(26.9)	10(17.2)		72(49.3)	7(70.0)
Clinical basic disease			.913			.084	NA
No	454(61.3)	36(62.1)		79(54.1)	4(40.0)
HE/HC/HL	138(18.6)	10(17.2)		30(20.5)	3(30.0)
Other	77(10.4)	5(8.6)		13(8.9)	3(30.0)
Mixed	72(9.7)	7(12.1)		24(16.4)	0(0.0)
Note.—Unless otherwise indicated, the data are qualitative variables, the number of patients are outside the brackets, the percentages are inside the brackets, and the statistical values are the Chi-square test results. N-LNM = non lymph node metastasis, LNM = lymph node metastasis, HL = hyperlipidemia, HE = hypertension, HC = hyperglycemia, NA = not applicable.
^*Data are means ± standard deviation, and the statistical values are the independent sample t-test results.

3.2. CT semantic features of patients

In the univariate analysis of CT semantic features, it showed difference in terms of pulmonary disease, solid component proportion, spiculation, air cavity, air bronchogram type, pleural contact type, mean diameter and CT_mean value between LNM group and N-LNM group (P＜0.05); but not in location, shape, and lobulation between the two groups (P＞0.05). After adjusting for confounding factors by multivariate logistic regression, pulmonary disease (P = 0.011), solid component proportion (P＜0.001), pleural contact type (P = 0.026), and mean diameter (P＜0.001) were shown as independent predictors of LNM (Table 2).

Table 2

CT semantic features of patients.
Variable	Univariate						Multivariate
	Training Group			Validation Group			Multivariate
	N-LNM	LNM	P	N-LNM	LNM	P	P value
Total cases	741(92.7)	58(7.3)		146(93.6)	10(6.4)
Pulmonary disease			＜.001^**			.311^**	.011
No	544(73.4)	46(79.3)		110(75.3)	6(60.0)
Emphysema/Bullae	57(7.7)	9(15.5)		18(12.3)	2(20.0)
Ventilation-perfusion imbalance	118(15.9)	0(0.0)		12(8.2)	1(10.0)
Mixed	22(3.0)	3(5.2)		6(4.1)	1(10.0)
Solid component proportion			＜.001			＜.001^**	＜.001
≤ 25%	442(59.6)	6(10.3)		72(49.3)	0(0.0)
≤ 50%	109(14.7)	6(10.3)		22(15.1)	1(10.0)
≤ 75%	95(12.8)	14(24.1)		23(15.8)	1(10.0)
＜100%	95(12.8)	32(55.2)		29(19.9)	8(80.0)
Location			.260			.846^**	NA
Right upper lobe	233(31.4)	20(34.5)		59(40.4)	6(60.0)
Right middle lobe	64(8.6)	4(6.9)		13(8.9)	0(0.0)
Right lower lobe	156(21.1)	6(10.3)		27(18.5)	1(10.0)
Left upper lobe	196(26.5)	21(36.2)		25(17.1)	2(20.0)
Left lower lobe	92(12.4)	7(12.1)		22(15.1)	1(10.0)
Shape			1.000^**			1.000^**	NA
Round	8(1.1)	0(0.0)		1(0.7)	0(0.0)
Oval	34(4.6)	2(3.4)		12(8.2)	0(0.0)
Irregular	699(94.3)	56(96.6)		133(91.1)	10(100.0)
Lobulation			.935			.018	NA
No	118(15.9)	9(15.5)		79(54.1)	1(10.0)
Yes	623(84.1)	49(84.5)		67(45.9)	9(90.0)
Spiculation			＜.001			.241	.133
No	397(53.6)	16(27.6)		118(80.8)	6(60.0)
Yes	344(46.4)	42(72.4)		28(19.2)	4(40.0)
Air cavity			.002^**			.071^**	.504
No	573(77.3)	38(65.5)		107(73.3)	5(50.0)
Vacuole	104(14.0)	13(22.4)		27(18.5)	5(50.0)
Cavity	62(8.4)	4(6.9)		12(8.2)	0(0.0)
Cavum	2(0.3)	3(5.2)		0(0.0)	0(0.0)
Air bronchogram type			.014^**			.162^**	.312
No	546(73.7)	37(63.8)		102(69.9)	6(60.0)
Without bronchial deformation	27(3.6)	8(13.8)		20(13.7)	0(0.0)
With bronchial deformation	128(17.3)	10(17.2)		23(15.8)	4(40.0)
Adjacent bronchiectasis	40(5.4)	3(5.2)		1(0.7)	0(0.0)
Pleural contact type			＜.001			.058^**	.026
No contact	172(23.2)	6(10.3)		35(24.0)	0(0.0)
Direct contact	341(46.0)	17(29.3)		38(26.0)	4(40.0)
Pleural traction without indentation	130(17.5)	5(8.6)		32(21.9)	1(10.0)
Pleural traction with indentation	32(4.3)	14(24.1)		14(9.6)	0(0.0)
Direct contact and traction	66(8.9)	16(27.6)		27(18.5)	5(50.0)
Mean diameter^***	16.0(12.0, 20.5)	23.2(16.2, 30.6)	＜.001	15.0(11.0,22.1)	31.7(22.4,37.6)	＜.001	＜.001
CT_mean value	19.3 ± 98.3	50.6 ± 26.1	＜.001^*	34.6(17.4,60.6)	32.1(21.5,44.6)	.764^***	.522
Note.—Unless otherwise indicated, the data are qualitative variables, the number of patients are outside the brackets, the percentages are inside the brackets, and the statistical values are the Chi-square test results. N-LNM = non lymph node metastasis, LNM = lymph node metastasis, CT_mean = mean of CT attenuation.
^*Data are means ± standard deviation, and the statistical values are the independent sample t-test results.
^**The statistical values are the results of Fisher's exact probability test.
^***Data do not conform to normal distribution, the median are outside parentheses, the lower quartile and the upper quartile are in parentheses, and the statistical values are Mann-Whitney U test results.

3.3. Development and comparison of prediction models

Combined model 1 was established based on the 9 risk factors that were statistically different in univariate analysis, and the combined model 2 included all independent risk factors. Pulmonary disease, solid component proportion, pleural contact type and mean diameter were fused as CT semantic features among independent predictors into the third combined model, named image predictors model. Regarding the performance of the three combined models and individual models (Table 3), in training group, the combined model 1 yielded an AUC of 0.880 (95% CI:0.777–0.891), with sensitivity of 81.2%, specificity of 78.5% and accuracy of 78.7%; the combined model 2 had an AUC of 0.876 (95% CI:0.796–0.892), with sensitivity of 77.0%, specificity of 77.6% and accuracy of 77.5%; the image predictors model yielded an AUC of 0.868 (95% CI:0.807–0.889), with sensitivity of 80.4%, specificity of 78.4% and accuracy of 78.6%. Solid component proportion was the single model with the best ROC among the four individual models, with an AUC of 0.815 (95% CI:0.734–0.844), sensitivity, specificity and accuracy of 80.9%, 72.7% and 73.3%, respectively; in validation group, the combined model 1 yielded an AUC of 0.853 (95% CI:0.777–0.891), with sensitivity of 72.9%, specificity of 77.8% and accuracy of 77.5%; the combined model 2 had an AUC of 0.861 (95% CI:0.796–0.893), with sensitivity of 76.0%, specificity and accuracy were both 77.2%; the image predictors model yielded an AUC of 0.857 (95% CI:0.807–0.888), and the sensitivity, specificity and accuracy were all 77.6%. The results of Delong test showed that there was no significant difference in diagnostic efficiency among the three combined models (all P＞0.05), but the solid component proportion was significantly from the performance of the three combined models (all P＜0.01). The ROC curves of the seven models in the training and validation groups are depicted in Fig. 4.

Table 3

Comparison of diagnostic efficacy of models.
Model Type	Threshold	AUC	95% CI	Sensitivity(%)	Specificity(%)	Accuracy(%)	P Value
Training Group
Combined model 1	0.532	0.880	0.777–0.891	81.2	78.5	78.7	NA
Combined model 2	0.462	0.876	0.796–0.892	77.0	77.6	77.5	NA
Image predictors model	0.462	0.868	0.807–0.889	80.4	78.4	78.6	NA
Pulmonary disease	0.480	0.521	0.423–0.557	76.5	26.3	29.8	NA
Solid component proportion	0.607	0.815	0.734–0.844	80.9	72.7	73.3	NA
Pleural contact type	0.653	0.687	0.654–0.714	60.3	66.0	65.6	NA
Mean diameter	0.531	0.736	0.725–0.737	61.1	78.4	73.8	NA
Validation Group
Combined model 1	0.563	0.853	0.777–0.891	72.9	77.8	77.5	0.2334^*
Combined model 2	0.554	0.861	0.796–0.893	76.0	77.2	77.2	0.5177^**
Image predictors model	0.575	0.857	0.807–0.888	77.6	77.6	77.6	0.7995^***
Pulmonary disease	0.799	0.521	0.423–0.557	76.4	26.3	29.8	0.0026^#
Solid component proportion	0.504	0.814	0.788–0.835	81.0	72.7	73.3	0.0006^#
Pleural contact type	0.622	0.713	0.674–0.753	59.3	66.0	65.7	0.0032^#
Mean diameter	0.577	0.736	0.721–0.747	63.3	74.1	73.3	NA
Note.—Combined Model 1 was established by 9 factors that were statistically different in the results of univariate analysis; Combined Model 2 included all independent risk factors from multivariate analysis; Image Predictors Model involved all imaging independent predictors [Pulmonary disease & Solid component proportion & Pleural contact type & Mean diameter].
^*Data is the result of Delong test between Combined model 1 and Combined model 2.
^**Data is the result of Delong test between Combined model 2 and Image predictors model.
^***Data is the result of Delong test between Image predictors model and Combined model 1.
^#Data are the results of Delong test of Solid component proportion in turn with Combined model 1, Combined model 2 and Image predictors model.

In this retrospective study, we analyzed the risk factors of LNM in 955 NSCLC patients manifested as part-solid nodules on CT findings from multiple centers. Of all the lymph nodes detected (9933) in training group, only 217 (2.2%) were cancerous, indicating a low incidence of LNM in part-solid NSCLC. According to the distribution of the examined lymph nodes in each station, it was found in this study that LNM occurred at stations 2–4 (superior mediastinal) with the highest probability, followed by the lung parenchyma, and no LNM occurred in the supraclavicular region in station 1; for N-LNM, the lymph nodes were mainly located in the N1 region, followed by the superior mediastinal region, and few lymph nodes were detected in the supraclavicular region and aortic region.

Regarding CT semantic features, the solid component proportion was an independent risk factor of LNM. In the present study, the ratio of solid component of nodule to tumor in the LNM group was mainly higher than 75% (32/53); therefore, the solid component of nodule was an adverse factor of LNM, which also been confirmed in other studies [22, 23]. Moreover, the mean value of CT attenuation (CT_mean) in the LNM group (50.6 ± 26.1) was significantly higher than that in the N-LNM group (19.3 ± 98.3), which we hypothesized that this was related to the solid component proportion of nodules; there was a positive correlation between solid component and CT attenuation value, and most nodules in the LNM group had a larger proportion of solid component. Perhaps this was one of reasons why the mean value of CT lost its significance in multivariate logistic regression. In terms of pleural contact type, a study of 478 peripheral solid NSCLC patients with a largest short-diameter > 5mm found that type II pleural involvement (a linear or cord-like pleural tag or tumor abut to the pleura with a broad base observed on both lung and mediastinal window images) was an independent predictor of occult lymph node metastasis (OLNM) [24]. Similar to them, in the present study, nodules were more likely to come into direct contact with the pleural than no contact, both in patients with LNM and N-LNM. However, for patients with LNM, pleural tags were more likely to cause indentation, and the probability of directing contact with the pleural concomitant pleural tags was also higher than that of N-LNM (27.6% vs. 8.9%). There are abundant lymphatic vessels on the surface of the subpleural lymphatic plexus [25]. We inferred that when the lesion is in contact with the pleura, the tumor cells first enter the lymphatic vessels, proliferate with lymph fluid in the lymphatic circulation, and enter the lymph node station to complete the invasion.

Emphysema is characterized pathologically by the presence of diffuse chronic inflammation of the lung parenchyma, oxidative stress, and lung destruction [26]. A previous study reported [27], emphysema was an independent risk factor for lung cancer, and the more severe the emphysema was, the higher the incidence was. The correlation between emphysema and LNM was also explored in the results of this study. When it comes to pulmonary disease, the results showed that patients with LNM were more likely to suffer from concomitant emphysema or bullae than patients with N-LNM. Some studies reported [28, 29], A longer mean diameter was an independent risk factor for LNM. DuComb et al. retrospectively studied 332 patients with T1 non-small cell lung cancer and showed that among patients with LNM, the most common pathological type was adenocarcinoma, which was consistent with the results of this study (56/58). In addition, they found that neither tumor diameter nor location was a risk factor for LNM [30]. In this study, location was also not shown to be a significant factor for metastases, but in contrast, mean diameter was an independent risk factor for the occurrence of LNM, possibly due to different inclusion and exclusion criteria: Our study included no restriction on tumor size, while they enrolled patients with T1 (8-30mm).

Previous studies have shown that model-based schemes can make better utilize radiographic information to predict lymph node diseases [31]. Das et al. integrated clinical parameters and radiomics features extracted from three ROIs of gross tumor volume (GTV), peritumoral volume (PTV), and LN in different ways to create different nomograms for predicting preoperative LNM in adenocarcinoma, and compared the predictive efficiency of each model [32]. The results showed that the AUC of radiological features based on GTV, PTV and LN in the external verification cohort were 0.74,0.72 and 0.64 respectively; the AUC of integrating GTV and PTV (GPTV) was 0.75 in the external validation cohort. GPTV combined with LN showed an AUC of 0.76, and the strongest predictive power was the integrated nomogram of clinical parameters along with CT radiomics information from GTV, PTV and LN, with an AUC of 0.79 (95%CI 0.66–0.93). Our study explored the predictive ability of three combined models to identify LNM. The results showed that compared with the combined model 1 including 9 risk factors and the combined model 2 based on all independent factors, the image predictors model established by pulmonary disease, solid component proportion, pleural contact type and mean diameter had the highest diagnostic accuracy in validation group, and its efficiency was better than all the individual models. The image predictors model only incorporated CT semantic features of the significant independent factors in the multivariate logistic regression results; this model embodied fewer fusion factors but higher predictive ability, which had profound implications: CT semantic features have a good ability to predict LNM in patients with NSCLC, and can provide preoperative guidance for clinical practice. Due to the poor image quality caused by respiratory movement and heartbeat, the prediction of LNM is limited in lung MR imaging [33]. PET is another imaging method for mediastinal staging, but its high cost, relatively low rate of lymph node involvement, and high false negative rate have hindered its routine implementation [34]. Martinez-Zayas et al. conducted a multicenter prospective validation of two retrospectively developed diagnostic models, called HAL and HOMER. In their study, they took 1799 NSCLC patients who underwent EBUS-TBNA staging and PET-CT as subjects. HOMER was used to predict N0 vs N1 vs N2/3 (three-classifier) and HAL was used to predict N2/3 vs N0/1 (two-classifier). Their results showed that HAL and HOMER had good multicentre discrimination: HAL had an AUC of 0.873, HOMER yielded AUC of 0.837 for predicting N1/2/3, and AUC of 0.876 for predicting N2/3 [35].

Several potential limitations of this study merit comment: 1) As its retrospective nature, potential selection bias would hinder the comparability and reproducibility of the results. 2) In the evaluation of CT image features, this study mainly focused on the characteristics of nodules. The image evaluation of lymph nodes can be considered to be added in later research for more in-depth exploration. 3) What we have to admit was that because this study analyzed image features based on subjective description and measurement, which restricted the consistency of models based on semantic features. With the development of radiomics that makes it possible to transform images into image quantitative feature data, it will contribute to describe tumor characteristics more objectively and quantitatively, so it is expected to have more studies to explore the prognosis prediction and survival analysis of LNM by radiomics.

In conclusion, smoking history, pulmonary disease, solid component proportion, pleural contact type, and mean diameter are independent risk factors for lymph node metastasis in non-small cell lung cancer patients with part-solid nodules. Furthermore, the image predictors model proposed in this study has a benign diagnostic efficiency for LNM, indicating that CT semantic features can guide clinical practice to some extent before operation, and has promising application value.

LNM Lymph Node Metastasis

NSCLC Non-Small Cell Lung Cancer

CT Computed Tomography

AUC Area Under the Curve

ROC Receiver Operating Characteristic Curve

TNM Tumor-Node-Metastasis

PSN Part-Solid Nodule

GGO Ground-Glass Opacity

FOV Field of View

ROI Region of Interest

GTV Gross Tumor Volume

PTV Peritumoral Volume

Ethics approval and consent to participate

This retrospective study was approved by the Ethics Committee of Tongde Hospital of Zhejiang Province (No.2022-029-JY) and the Ethical Committee of the First Affiliated Hospital of Bengbu Medical College (No.2021-292), and waived the need of informed consent. The study was conducted in accordance with the Declaration of Helsinki.

Consent for publication

Not applicable.

Availability of data and materials

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

Competing interests

The authors declare no competing interests.

Funding

This research was supported by Medical Health Science and Technology Project of Zhejiang Province (Grant No.2021KY602 and 2022KY702) and Key Research and Development Program of Anhui Province (Grant No.2022e07020033).

Authors' contributions

Yang Yang: Conceptualization, Methodology, Formal analysis, Investigation, Data Curation, Writing - Original Draft. Jian Wang: Conceptualization, Methodology, Formal analysis, Investigation, Resources, Data Curation, Writing - Review & Editing, Supervision, Funding acquisition. Zhongfeng Niu: Methodology, Formal analysis, Investigation, Resources, Data Curation, Writing - Review & Editing. Guoqun Mao: Methodology, Formal analysis, Investigation, Resources, Data Curation, Writing - Review & Editing. Xiandi Zhu: Methodology, Formal analysis, Investigation, Writing - Review & Editing. Zhihua Xu: Methodology, Formal analysis, Investigation, Writing - Review & Editing. Dengfa Yang: Methodology, Formal analysis, Investigation, Resources, Data Curation, Writing - Review & Editing. Hui Wang: Methodology, Formal analysis, Investigation, Data Curation. Maosheng Xu: Conceptualization, Methodology, Formal analysis, Investigation, Resources, Data Curation, Writing - Review & Editing. Zongyu Xie: Conceptualization, Methodology, Formal analysis, Investigation, Resources, Data Curation, Writing - Review & Editing, Supervision, Funding acquisition.

Acknowledgements

Not applicable.

H Sung, J Ferlay, RL Siegel, M Laversanne, I Soerjomataram, A Jemal, et al. Global cancer statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2021;71(3):209-249. https://doi.org/10.3322/caac.21660.
F Bray, J Ferlay, I Soerjomataram, RL Siegel, LA Torre, A Jemal. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2018;68(6):394-424. https://doi.org/10.3322/caac.21492.
D Morgensztern, SH Ng, F Gao, R Govindan. Trends in stage distribution for patients with non-small cell lung cancer: a national cancer database survey. J Thorac Oncol. 2010;5(1):29-33. https://doi.org/10.1097/JTO.0b013e3181c5920c.
RL Siegel, KD Miller, HE Fuchs, A Jemal. Cancer statistics, 2022. CA Cancer J Clin. 2022;72(1):7-33. https://doi.org/10.3322/caac.21708.
WF Tang, M Wu, H Bao, Y Xu, JS Lin, Y Liang, et al. Timing and origins of local and distant metastases in lung cancer. J Thorac Oncol. 2021;16(7);1136-1148. https://doi.org/10.1016/j.jtho.2021.02.023.
H Asamura, K Chansky, J Crowley, P Goldstraw, VW Rusch, JF Vansteenkiste, et al. The international association for the study of lung cancer lung cancer staging project: proposals for the revision of the N descriptors in the forthcoming 8th edition of the TNM classification for lung cancer. J Thorac Oncol. 2015;10(12):1675-1684. https://doi.org/10.1097/JTO.0000000000000678.
R Rami-Porta, H Asamura, WD Travis, VW Rusch. Lung cancer - major changes in the American Joint Committee on Cancer eighth edition cancer staging manual. CA Cancer J Clin. 2017;67(2):138-155. https://doi.org/10.3322/caac.21390.
S Watanabe, H Asamura. Lymph node dissection for lung cancer: significance, strategy, and technique. J Thorac Oncol. 2009;4(5):652-657. https://doi.org/10.1097/JTO.0b013e31819cce50.
WZ Zhong, SY Liu, YL Wu. Numbers or stations: from systematic sampling to individualized lymph node dissection in non-small-cell lung cancer. J Clin Oncol. 2017;35(11):1143-1145. https://doi.org/10.1200/JCO.2016.71.8544.
HJ de Koning, CM van der Aalst, PA de Jong, ET Scholten, K Nackaerts, MA Heuvelmans, et al. Reduced lung-cancer mortality with volume CT screening in a randomized trial. N Engl J Med. 2020;382(6):503-513. https://doi.org/10.1056/NEJMoa1911793.
DR Aberle, AM Adams, CD Berg, WC Black, JD Clapp, RM Fagerstrom, et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N Engl J Med. 2011;365(5):395-409. https://doi.org/10.1056/NEJMoa1102873.
G Wu, HC Woodruff, J Shen, T Refaee, S Sanduleanu, A Ibrahim, et al. Diagnosis of invasive lung adenocarcinoma based on chest CT radiomic features of part-solid pulmonary nodules: A multicenter study. Radiology. 2020;297(2):451-458. https://doi.org/10.1148/radiol.2020192431.
M Yanagawa, M Tsubamoto, Y Satoh, A Hata, T Miyata, Y Yoshida, et al. Lung adenocarcinoma at CT with 0.25-mm section thickness and a 2048 matrix: high-spatial-resolution imaging for predicting invasiveness. Radiology. 2020;297(2):462-471. https://doi.org/10.1148/radiol.2020201911.
AA Bankier, H MacMahon, JM Goo, GD Rubin, CM Schaefer-Prokop, DP Naidich. Recommendations for measuring pulmonary nodules at CT: a statement from the Fleischner Society. Radiology. 2017;285(2):584-600. https://doi.org/10.1148/radiol.2017162894.
KG Tournoy, JT Annema, M Krasnik, FJF Herth, JP van Meerbeeck. Endoscopic and endobronchial ultrasonography according to the proposed lymph node map definition in the seventh edition of the tumor, node, metastasis classification for lung cancer. J Thorac Oncol. 2009;4(12):1576-1584. https://doi.org/10.1097/JTO.0b013e3181c1274f.
MPJ Visser, I van Grimbergen, J Hölters, WB Barendregt, LC Vermeer, W Vreuls, et al. Performance insights of endobronchial ultrasonography (EBUS) and mediastinoscopy for mediastinal lymph node staging in lung cancer. Lung Cancer. 2021;156:122-128. https://doi.org/10.1016/j.lungcan.2021.04.003.
AH El-Sherief, CT Lau, BW Carter, CC Wu. Staging lung cancer: regional lymph node classification. Radiol Clin North Am. 2018;56(3):399-409. https://doi.org/10.1016/j.rcl.2018.01.008.
B Ye, M Cheng, W Li, XX Ge, JF Geng, J Feng, et al. Predictive factors for lymph node metastasis in clinical stage IA lung adenocarcinoma. Ann Thorac Surg. 2014;98(1):217-223. https://doi.org/10.1016/j.athoracsur.2014.03.005.
G Yang, P Nie, L Zhao, J Guo, W Xue, L Yan, et al. 2D and 3D texture analysis to predict lymphovascular invasion in lung adenocarcinoma. Eur J Radiol. 2020;129:109111. https://doi.org/10.1016/j.ejrad.2020.109111.
CV Zwirewich, S Vedal, RR Miller, NL Müller. Solitary pulmonary nodule: high-resolution CT and radiologic-pathologic correlation. Radiology. 1991;179(2):469-476. https://doi.org/10.1148/radiology.179.2.2014294.
JS Hsu, IT Han, TH Tsai, SF Lin, TS Jaw, GC Liu, et al. Pleural tags on CT scans to predict visceral pleural invasion of non-small cell lung cancer that does not abut the pleura. Radiology. 2016;279(2):590-596. https://doi.org/10.1148/radiol.2015151120.
W Li, F Zhou, Z Wan, M Li, Y Zhang, X Bao, et al. Clinicopathologic features and lymph node metastatic characteristics in patients with adenocarcinoma manifesting as part-solid nodule exceeding 3 cm in diameter. Lung Cancer. 2019;136:37-44. https://doi.org/10.1016/j.lungcan.2019.07.029.
JY Cho, CS Leem, Y Kim, ES Kim, SH Lee, YJ Lee, et al. Solid part size is an important predictor of nodal metastasis in lung cancer with a subsolid tumor. BMC Pulm Med. 2018;18(1):151. https://doi.org/10.1186/s12890-018-0709-2.
XQ He, TY Luo, X Li, JW Huo, JW Gong, Q Li. Clinicopathological and computed tomographic features associated with occult lymph node metastasis in patients with peripheral solid non-small cell lung cancer. Eur J Radiol. 2021;144:109981. https://doi.org/10.1016/j.ejrad.2021.109981.
E Lee, DM Biko, W Sherk, WR Masch, M Ladino-Torres, PP Agarwal. Understanding lymphatic anatomy and abnormalities at imaging. Radiographics. 2022;42(2):487-505. https://doi.org/10.1148/rg.210104.
X Yang, HJ Wisselink, R Vliegenthart, MA Heuvelmans, HJM Groen, M Vonder, et al. Association between chest CT-defined emphysema and lung cancer: a systematic review and meta-analysis. Radiology. 2022;304(2):322-330. https://doi.org/10.1148/radiol.212904.
AR Hunsaker. Emphysema as a predictor of lung cancer: implications for lung cancer screening. Radiology. 2022;304(2):331-332. https://doi.org/10.1148/radiol.220697.
L Wang, W Jiang, C Zhan, Y Shi, Y Zhang, Z Lin, et al. Lymph node metastasis in clinical stage IA peripheral lung cancer. Lung Cancer. 2015;90(1):41-46. https://doi.org/10.1016/j.lungcan.2015.07.003.
Y Zhong, M Yuan, T Zhang, YD Zhang, H Li, TF Yu. Radiomics approach to prediction of occult mediastinal lymph node metastasis of lung adenocarcinoma. AJR Am J Roentgenol. 2018;211(1):109-113. https://doi.org/10.2214/AJR.17.19074.
EA DuComb, BA Tonelli, Y Tuo, BF Cole, V Mori, JHT Bates, et al. Evidence for expanding invasive mediastinal staging for peripheral T1 lung tumors. Chest. 2020;158(5):2192-2199. https://doi.org/10.1016/j.chest.2020.05.607.
FC Verdial, DK Madtes, B Hwang, MS Mulligan, K Odem-Davis, R Waworuntu, et al. Prediction model for nodal disease among patients with non-small cell lung cancer. Ann Thorac Surg. 2019;107(6):1600-1606. https://doi.org/10.1016/j.athoracsur.2018.12.041.
SK Das, KW Fang, L Xu, B Li, X Zhang, HF Yang. Integrative nomogram of intratumoral, peritumoral, and lymph node radiomic features for prediction of lymph node metastasis in cT1N0M0 lung adenocarcinomas. Sci Rep. 2021;11(1):10829. https://doi.org/10.1038/s41598-021-90367-4.
J Peerlings, EGC Troost, PJ Nelemans, DCP Cobben, JCJ de Boer, AL Hoffmann, et al. The diagnostic value of MR imaging in determining the lymph node status of patients with non-small cell lung cancer: a meta-analysis. Radiology. 2016;281(1):86-98. https://doi.org/10.1148/radiol.2016151631.
Y Zhong, Y She, J Deng, S Chen, T Wang, M Yang, et al. Deep learning for prediction of N2 metastasis and survival for clinical stage I non-small cell lung cancer. Radiology. 2022;302(1):200-211. https://doi.org/10.1148/radiol.2021210902.
G Martinez-Zayas, FA Almeida, L Yarmus, D Steinfort, DR Lazarus, MJ Simoff, et al. Predicting lymph node metastasis in non-small cell lung cancer: prospective external and temporal validation of the HAL and HOMER models. Chest. 2021;160(3):1108-1120. https://doi.org/10.1016/j.chest.2021.04.048.

No competing interests reported.

Download PDF

Version 1

posted

You are reading this latest preprint version

Preoperative CT semantic features predict lymph node metastasis of part-solid nodules in non-small cell lung cancer: a multicenter retrospective study

Status:

Version 1

Abstract

Figures

1. Introduction

2. Material And Methods

2.1. Patients

2.2. CT image acquisition

2.3. Imaging evaluation

2.4. Statistical analysis

3. Results

3.1. Clinical characteristics of patients in training group

3.2. CT semantic features of patients

3.3. Development and comparison of prediction models

4. Discussion

5. Conclusions

Abbreviations

Declarations

References

Additional Declarations

Status:

Version 1