Sage Journals: Discover world-class research

Abstract

BACKGROUND:

Image-based biomarkers could have translational implications by characterizing tumor behavior of lung cancers diagnosed during lung cancer screening. In this study, peritumoral and intratumoral radiomics and volume doubling time (VDT) were used to identify high-risk subsets of lung patients diagnosed in lung cancer screening that are associated with poor survival outcomes.

METHODS:

Data and images were acquired from the National Lung Screening Trial. VDT was calculated between two consequent screening intervals approximately 1 year apart; peritumoral and intratumoral radiomics were extracted from the baseline screen. Overall survival (OS) was the main endpoint. Classification and Regression Tree analyses identified the most predictive covariates to classify patient outcomes.

RESULTS:

Decision tree analysis stratified patients into three risk-groups (low, intermediate, and high) based on VDT and one radiomic feature (compactness). High-risk patients had extremely poor survival outcomes (hazard ratio [HR] $=$ 8.15; 25% 5-year OS) versus low-risk patients (HR $=$ 1.00; 83.3% 5-year OS). Among early-stage lung cancers, high-risk patients had poor survival outcomes (HR $=$ 9.07; 44.4% 5-year OS) versus the low-risk group (HR $=$ 1.00; 90.9% 5-year OS). For VDT, the decision tree analysis identified a novel cut-point of 279 days and using this cut-point VDT alone discriminated between aggressive (HR $=$ 4.18; 45% 5-year OS) versus indolent/low-risk cancers (HR $=$ 1.00; 82.8% 5-year OS).

CONCLUSION:

We utilized peritumoral and intratumoral radiomic features and VDT to generate a model that identify a high-risk group of screen-detected lung cancers associated with poor survival outcomes. These vulnerable subset of screen-detected lung cancers may be candidates for more aggressive surveillance/follow-up and treatment, such as adjuvant therapy.

Keywords

Radiomics LDCT NLST CART early detection VDT

1. Introduction

Four randomized clinical trials demonstrated that screening and early detection of lung cancer by low-dose helical computed tomography (LDCT) reduces overall and lung cancer-specific mortality [1, 2, 6, 11, 33]. However, despite the benefits of early detection, cancer screening can result in overdiagnosis which is the detection of asymptomatic cancers that are non-growing or slow-growing and unlikely fatal if left untreated [10]. Overdiagnosis of pulmonary nodules can result in unnecessary diagnostic procedures that are often invasive, associated with increased costs, and associated with added stress for patients and their families [52]. In the US National Lung Screening Trial (NLST), 10 to 27% lung cancers were overdiagnosed in the LDCT arm of the study [27, 34, 41, 42, 43]. As such, biomarkers that can discriminate between the variability in growth rates and lethal potential [42] of lung cancers diagnosed in lung cancer screening would have important clinical translational implications.

Radiomics is the conversion of medical images into structured, mineable data that can subsequently be utilized for statistical analysis and modeling [9, 12, 13, 17, 25, 28, 29, 44]. Radiomic features have shown to provide rapid and accurate noninvasive biomarkers for lung cancer risk prediction, diagnostics, prognosis, treatment response monitoring, and tumor biology [15, 45, 47]. Radiomics have many advantages over tissue-based biomarkers as they are generated from standard-of-care imaging and they reflect the pathophysiology of the entire imaged tumor or region-of-interest (e.g., screen-detected nodule) and not just the portion of the tumor that was assayed in the case for tissue-based biomarkers.

Volume doubling time (VDT) of a pulmonary nodule, defined as the number of days in which the nodule doubles in volume, is a clinically relevant metric in lung cancer screening. VDT has clinical utility in discriminating between nodules that are infectious/inflammatory, malignant, and benign [30, 46, 51]. Generally, nodules that are of an infectious or inflammatory pathophysiology have a VDT of less than 20 days, a VDT of less than 400 days (and greater than 20 days) represents a high likelihood of malignancy, and a VDT above 500 days is likely a benign nodule [23, 24]. To date, there have been few studies analyzing VDT, a delta-radiomic, as a potential biomarker to evaluate lethal potential of lung cancers diagnosed in lung cancer screening. Thus, using publicly available data and LDCT images from the NLST, we generated radiomic features to determine if we can discriminate between indolent, slow, and aggressive incidentally diagnosed lung cancers. Radiomic features describing size, shape, volume, growth rate (i.e., VDT), and textural characteristics were calculated from the intratumoral region (area within the tumor) and from the peritumoral region (area surrounding the tumor parenchyma). The overarching goal of this study was to utilize radiomics biomarkers to generate parsimonious models to predict lung cancer outcomes in the lung cancer screening setting.

2. Materials and methods

2.1 NLST study population

Deidentified LDCT images and data were obtained from the National Cancer Institute (NCI) Cancer Data Access System (CDAS) [14]. The NLST study design and main findings have been described previously [1, 2]. NLST eligibility criteria included current and former smokers aged 55–74 years with a minimum 30 pack-years smoking history and former smokers had to have quit within the past 15 years.

The present study used a subset of lung cancers diagnosed in the LDCT-arm of the NLST and has been described elsewhere [38]. Briefly, we identified 314 screen-detected, incident lung cancer patients who were not diagnosed with lung cancer at baseline screening (T0) but were subsequently diagnosed with lung cancer either at the first follow-up screening (T1) or second follow-up screening (T2). However, participants were excluded due to several reasons: unavailable complete volumetric images sets, nodules at baseline that could not be identified, and nodules where it was difficult to contour the tumor. Lung cancer patients who had multiple nodules at time of their diagnosis were excluded ( $N=$ 20) since we are unable to verify which nodule(s) were cancer.

The final analytical cohort included 94 participants who had a negative or positive baseline LDCT screening. Based on NLST criteria, positive screens were defined as abnormalities on baseline screens or at follow-up screens that were new, stable or evolved that resulted as one or more non-calcified masses measuring $\geqslant$ 4 mm in axial diameter [38]. Negative screens were defined as having no abnormalities, minor abnormalities, or significant abnormalities not suspicious for lung cancer.

Figure 1.

Schema for identification of patients diagnosed with cancer at two different timepoints in the NLST.

2.2 Radiomics

Nodule identification and tumor segmentation has been described in detail elsewhere [17]. Briefly, the tumor mask images (i.e., tumor delineations) were imported into in-house radiomic feature extraction toolboxes created in MATLAB ${}^{\@setsize{\scriptsize}{8pt}{\viipt}{\@viipt}\textregistered}$ 2015b (The Mathworks Inc., Natick, Massachusetts) and C $++$ (https://isocpp. org). Using cubic interpolation, the images were resampled to a single voxel spacing of 1 mm $\times$ 1 mm $\times$ 1 mm to standardize spacing across all images. Hounsfield units (HU) in all images were resampled into fixed bin sizes of 25 HUs discretized from $-$ 1000 to 1000 HU.

Using standardized radiomic algorithms from the Image Biomarker Standardization Initiative (IBSI) v5 [56], a total of 264 radiomic features were extracted from the semi-automatically segmented intratumoral region ( $n=$ 155) and from the peritumoral region ( $n=$ 109) 3 mm outside of the tumor boundary. The peritumoral regions were generated as an extension of the tumor segmentations using morphological image processing operations as previously described [48]. Peritumoral regions were bounded by a lung parenchyma mask to exclude the region of interest (ROI) outside of the lung parenchyma. Shape- and size-based peritumoral features were excluded as they explicitly describe and correlate with the intratumoral ROI. The 65 most stable and reproducible intratumoral and peritumoral radiomic features that were previously found by our group [48] were utilized for analysis.

Volume doubling time was calculated for the screening interval from T0 to T1 and for the screening interval from T1 to T2 (Fig. 1). Below is the equation for VDT where Ti is the time interval between two LDCT imaging studies, $V_{i}$ is the volume of the nodule in the second LDCT image, and $V_{0}$ is the volume of nodule in the first LDCT image:

$\displaystyle\textit{VDT}=\frac{\ln 2xT_{i}}{\ln\left({\frac{V_{i}}{V_{0}}}% \right)}$

2.3 Radiogenomics analysis

A previously described dataset [39] of surgically resected adenocarcinoma lung cancers who had pre-surgery CTs and gene expression data was used to identify potential biological underpinnings of the most informative radiomic features identified in our analyses. The gene probesets were filtered and determined as significant using the following criteria based on a Student’s $t$ test $p<$ 0.001 and mean log fold-change between high and low prognostic radiomic feature oflfc $>$ log2 (1.4 FC). The significant probesets from the analyses were intersected yielding a final list of probesets significantly associated with the most informative radiomic features. ANOVA and Tukey pairwise mean comparison was performed to analyzed gene expression across the risk groups.

2.4 Statistical analysis

Statistical analyses were performed using Stata/MP 14.2 (StataCorp LP, College Station TX), R Project for Statistical Computing (version 3.5.2), and R Studio (version 1.1.463).

Overall survival (OS) and progression-free survival (PFS) were the main endpoints and were assessed from date of lung cancer diagnosis to the date of an event or last follow up. For OS, an event was defined as death and for PFS an event was established as death or progression of cancer. All survival data were right censored at 5-years.

To generate a parsimonious model, the 65 stable radiomic features and VDT were subjected to Classification and Regression Tree (CART) analysis, adapted for survival analysis, to stratify patients into risk groups. The risk groups were analyzed by Cox regression models, Kaplan-Meier survival curves, and log-rank tests. The most informative radiomic features identified in the CART analysis were internally cross-validated using R package “rms” [16]. The Harrell’s concordance index (C-index) was used to evaluate the performance of the multivariable Cox regression models. Backward logistic regression model was used to identify radiomics features significantly associated with VDT dichotomized.

3. Results

3.1 Patient characteristics

The demographics of the lung cancer patients are provided in Table 1. Mean age of enrollment into the NLST was 64.3 years, 58.5% participants were male, 58.5% were current smokers, 55.3% were diagnosed with adenocarcinoma, and 80.8% had an early-stage lung cancer (stage I or II).

Table 1
Patient characteristics

Characteristics	All patients ( $N=$ 94)
Age, mean(sd)	64.3	(5.4)
Sex
Female	39	(41.5%)
Male	55	(58.5%)
Smoking status
Former	39	(41.5%)
Current	55	(58.5%)
Pack-years smoked mean (sd)	63.4	(23.9)
Self-reported COPD
No	81	(86.2%)
Yes	13	(13.8%)
Family history of lung cancer
No	72	(76.6%)
Yes	22	(23.4%)
Histology
Adenocarcinoma-BAC	52	(55.3%)
Small cell	3	(3.2%)
Squamous	18	(19.2%)
Other NOS	21	(22.3%)
Stage
I and II	76	(80.8%)
III and IV	18	(19.2%)
Treatment
Surgical	74	(78.7%)
Chemotherapy/other	6	(6.4%)
Radiation	14	(14.9%)
Baseline screening (T0)
Positive (T0 $+$ )	40	(42.6%)
Negative (T0 $-$ )	54	(57.4%)

Figure 2.

Risk-groups associated with overall survival for all patients and among early-stage patients diagnosed in the screening interval T0 to T1. (A) Schema identifying patients diagnosed in the screening interval T0 to T1. (B) The tree structure from the classification and regression tree (CART) analysis identified four risk groups based on two radiomics features and VDT. VDT was statistically significant different between very-high, high, and intermediate when compared to low-risk group. (C) Overall survival for VDT dichotomized by 234 days. Overall survival for the risk patient risk groups among all patients (D) and for early-stage (E), and for late-stage patients (F).

Figure 3.

Risk-groups associated with overall survival for all patients and among early-stage patients diagnosed in the screening interval T0 to T1 or T1 to T2. (A) Schema including patients diagnosed in the screening interval T1 to T2. (B) The tree structure from the classification and regression tree analysis (CART) identified three risk groups based on one radiomics feature and VDT. VDT was statistically significant different between high- and intermediate risk groups when compared to low-risk group. (C) Overall survival for VDT dichotomized by 279 days. Overall survival for the risk patient risk groups among all patients (D) and for early-stage (E), and for late-stage patients (F).

3.2 Screening interval from T0 to T1

Radiomics were calculated from LDCT images at T0 and VDT was calculated from T0 to T1 as seen in Fig. 2A. The mean number of days between T0 and T1 was 369 days (range: 282 to 620 days). VDT was not significantly correlated to intratumoral or peritumoral features. VDT and the 65 radiomic features that we identified as stable and reproducible were included as inputs into CART analysis and subsequent decision tree revealed four risk groups as terminal nodes: low-risk, intermediate-risk, high-risk, and very-high-risk (Fig. 2B). The decision tree structure included VDT and two radiomic features (intratumoral feature compactness and peritumoral feature average co-occurrence). There are differences in VDT across the four risk-groups. The very-high risk group has a lower VDT (median VDT: 108.56) between T0 to T1 when compared to the high, intermediate, and low risk group (median VDT: 114.6, 143.2, and 499.2, respectively). From T0 to T1, Kruskal-Wallis test revealed that there is a statistically significant difference in VDT between the four risk groups ( $P=$ 0.0001). Pairwise analyses revealed, that when compared to the low-risk group, VDT was statistically significant different for the intermediate-risk ( $P<$ 0.0001), high-risk group ( $P<$ 0.0001), and very-high risk group ( $P<$ 0.0001) as presented in Fig. 2B. CART analysis identified a novel VDT threshold (i.e., cutpoint) of 234 days which discriminated between high-risk lung cancers (HR $=$ 3.49; 40% 5-year OS) versus low-risk cancers (HR $=$ 1.00; 80.8% 5-year OS) (Fig. 2C). Similar results were observed for PFS(Supplemental Fig. 1A). The very-high-risk group (Fig. 2D) was associated with extremely poor OS (HR $=$ 11.71; 42.9% 2-year OS and 21.4 % 5-year OS, log-rank $P<$ 0.0001) versus the high (HR $=$ 5.08; 82.4% 2-year OS and 22.9% 5-year OS), intermediate (HR $=$ 1.61; 93.3% 2-year OS and 72.7% 5-year OS) and low-risk group (HR $=$ 1.00; 89.5% 2-year OS and 82.1% 5-year OS). Similar findings were observed for PFS(Supplemental Fig. 2A). Among early-stage patients, the very-high-risk group was associated with extremely poor survival (HR $=$ 14.84; 33.3% 2-year and 33.3% 5-year OS, log-rank $P<$ 0.001) versus the low-risk group (HR $=$ 1.00; 95.2% 2 year and 90.4% 5-year OS) (Fig. 2E). Among late-stage patients, there was no significant association with OS (Fig. 2F). Similar results were found for PFS (Supplemental Fig. 2B and C).

There were no statistically significant differences between the four risk groups by age, smoking status, number of pack-years smoked, baseline screening, and family history of lung cancer (Table 2). However, there were statistically significant differences across the risk groups for sex ( $P=$ 0.02), self-reported COPD ( $P=$ 0.003), histology ( $P=$ 0.007), treatment ( $P=$ 0.038), and stage of disease ( $P=$ 0.042). In terms of lung cancer stage, 64.3% of the patients in the very-high-risk group had lung cancer early-stage vs. 87.5% in the low-risk group ( $P=$ 0.042). As such, multivariable Cox regression models were used to adjust for potential confounding factors including sex, treatment, self-reported COPD, and stage. The very-high-risk group was associated with an elevated hazard ratio (OS: HR $=$ 10.90; 95% Confidence Interval: [3.98, 29.85] and PFS: HR $=$ 4.72; 95% Confidence Interval: [1.85, 12.07]) when compared to high, intermediate, and low-risk groups (Table 3 and Supplemental Table 1) with a C-index of 0.85. The multivariable model using 10-fold cross validation achieved high prediction performance (C-index $=$ 0.81). Among early-stage patients, the very-high-risk group was associated with an elevated hazard ratio (OS: HR $=$ 21.63; 95% Confidence Interval: [3.70, 123.60] and PFS: HR $=$ 5.92; 95% Confidence Interval: [1.81, 12.07]) when compared to high, intermediate and low-risk groups (Table 3 and Supplemental Table 1) with a C-index of 0.84.

3.3 Screening intervals from T0 to T1 and T1 to T2

For this analysis, we utilized the data from Section 3.2 above and included participants who were diagnosed with lung cancer at T2. Radiomics were calculated from T0 and VDT was calculated from T0 to T1 and between T1 to T2, as seen in Fig. 3A. The mean number of days between T1 and T2 was 370 days (range: 287 to 480 days). The same stable radiomic features were combined with VDT and subjected to CART analysis and revealed three risk groups: low-risk, intermediate-risk, high-risk (Fig. 3B). Due to the addition of participants who were diagnosed with lung cancer at T2, a new decision tree structure was identified. The decision tree structure included one radiomic feature (intratumoral feature compactness) and VDT. There are differences in VDT across the three risk-groups. The high-risk group has a lower VDT (median VDT: 104.58) between T0 to T1 and from T1 to T2 when compared to the intermediate, and low risk group (median VDT: 132.53 and 459.23, respectively) ( $P=$ 0.0001). Pairwise analyses demonstrated, that when compared to the low-risk group, median VDT was statistically significant different for intermediate-risk ( $P<$ 0.0001) and high-risk group ( $P<$ 0.0001) as presented Fig. 3B. CART identified a novel VDT threshold (VDT $=$ 279 days) that discriminated between high-risk lung cancers (HR $=$ 4.18; 45% 5-year OS) and low-risk lung cancers (HR $=$ 1.00; 82.8% 5-year OS) (Fig. 3C). Similar results were observed for PFS (Supplemental Fig. 1B). The high-risk group was associated with extremely poor OS (Hazard Ratio (HR) $=$ 8.15; 56.3% 2-year OS and 25.0 % 5-year OS, log-rank $P<$ 0.0001) versus the intermediate (HR $=$ 3.28; 78.0% 2-year OS and 53.3% 5-year OS), and low-risk group (HR $=$ 1.00; 100.0% 2-year OS and 83.3% 5-year OS) (Fig. 3D). Similar findings were observed for Progression Free Survival (PFS)(Supplemental Fig. 3A). Among early-stage patients, the high-risk group was associated with a significantly decreased OS (HR $=$ 9.07; 66.7% 2-year and 44.4% 5-year OS, log-rank $P<$ 0.0001) versus the low-risk group (HR $=$ 1.00; 96.9% 2-year and 90.9% 5-year OS) (Fig. 3E). When stratifying by late-stage patients, there was no significant association with OS (Fig. 3F). Similar results were found for PFS(Supplemental Fig. 3B and C).

There were no statistically significant differences between the three risk groups by age, smoking status, number of pack-years smoked, treatment, baseline screening, and family history of lung cancer (Table 2). However, there were statistically significant differences across the risk groups for sex ( $P=$ 0.003), self-reported COPD ( $P=$ 0.003), histology ( $P=<$ 0.0001), and stage of disease ( $P=$ 0.018). In term of lung cancer stage, 56.3% of the patients in the high-risk group had lung cancer early-stage vs. 89.2% in the low-risk group ( $P=$ 0.018). As such, multivariable Cox regression models were used to adjust for potential confounding factors including sex, treatment, and stage. The high-risk group was associated with an elevated hazard ratio (OS: HR $=$ 3.62; 95% CI: [1.12, 11.69] and PFS: HR $=$ 2.33; 95% CI: [0.90, 6.05]) when compared to intermediate, and low-risk groups (Table 3 and Supplemental Table 1) with a C-index of 0.78 in OS. Among early-stage patients, the high-risk group was associated with an elevated hazard ratio (OS: HR $=$ 7.82; 95% Confidence Interval: [1.44, 42.43] and PFS: HR $=$ 3.30; 95% Confidence Interval: [0.88, 12.35]) when compared to intermediate and low-risk groups (Table 3 and Supplemental Table 1) with a C-index of 0.72 in OS.

3.4 Predicting VDT using T0 radiomics features

When dichotomizing VDT by the novel threshold of 279 days, backward logistic regression identified the intratumoral radiomic feature compactness as the most informative radiomic feature (OR: 2.54; 95% CI 0.91, 7.12]). Compactness 2, a morphological feature, is a measure of how compact the shape of the region-of-interest is relative to a sphere (most compact) [3]. Compactness 2 was extracted from the semi-automatically segmented intratumoral region and peritumoral 3 mm outside of tumor boundary. Intratumoral Compactness 2 was identified as a stable and reproducible radiomic feature [48].

Table 2
Patient characteristics by risk group

	Screening interval from T0 to T1									Screening interval from T0 to T1 or T1 to T2
Characteristics	Low risk group ( $N=$ 48)		Intermediate risk group ( $N=$ 15)		High risk group ( $N=$ 17)		Very high risk group ( $N=$ 14)		$P$ -value ${}^{1,2,3}$	Low risk group ( $N=$ 37)		Intermediate risk group ( $N=$ 41)		High risk group ( $N=$ 16)		$P$ -value ${}^{1,2,3}$
Age, mean (sd)	62.6	(5.2)	64.4	(5.0)	65.2	(6.1)	68.3	(3.4)	0.13	63.7	(5.3)	63.8	(5.4)	66.6	(5.0)	0.93
Sex, N (%)									0.02							0.0003
Female	27	(56.3%)	4	(26.7%)	5	(29.4%)	3	(21.4%)		23	(62.2%)	10	(24.4%)	6	(37.5%)
Male	21	(43.7%)	11	(73.3%)	12	(70.6%)	11	(78.6%)		14	(37.8%)	31	(75.6%)	10	(62.5%)
Smoking status, $N$ (%)									0.60							0.93
No	20	(41.7%)	8	(53.3%)	5	(29.4%)	6	(42.9%)		16	(43.2%)	17	(41.5%)	6	(37.5%)
Yes	28	(58.3%)	7	(46.7%)	12	(70.6%)	8	(57.1%)		21	(56.8%)	24	(58.5%)	10	(62.5%)
Pack-years, mean (sd)	60.1	(19.5)	72.4	(33.3)	66.7	(25.2)	63.9	(24.7)	0.06	62.7	(19.8)	62.8	(26.9)	66.7	(26.2)	0.16
FH of lung cancer, $N$ (%)									0.15							0.39
No	40	(83.3%)	10	(66.7%)	14	(82.4%)	8	(57.1%)		31	(83.8%)	30	(73.2%)	11	(68.8%)
Yes	8	(16.7%)	5	(33.3%)	3	(17.6%)	6	(42.9%)		6	(16.2%)	11	(26.8%)	5	(31.3%)
Self-reported history of COPD, $N$ (%)									0.003							0.003
No	45	(93.8%)	14	(93.3%)	10	(58.8%)	12	(85.7%)		36	(97.3%)	35	(85.4%)	10	(62.5%)
Yes	3	(6.2%)	1	(6.7%)	7	(41.2%)	2	(14.3%)		1	(2.7%)	6	(14.6%)	6	(37.3%)
Histology, $N$ (%)									0.007							$<$ 0.0001
Adenocarcinoma-BAC ${}^{4}$	30	(62.5%)	7	(46.7%)	7	(41.2%)	8	(57.1%)		29	(78.4%)	18	(43.9%)	5	(31.3%)
Squamous	6	(12.5%)	7	(46.7%)	2	(11.8%)	3	(21.4%)		3	(8.1%)	12	(29.3%)	3	(18.8%)
Small cell	0	(0%)	0	(0%)	1	(5.9%)	2	(14.3%)		0	(0%)	0	(0%)	3	(18.8%)
Other NOS	12	(25%)	1	(6.7%)	7	(41.2%)	1	(7.1%)		5	(13.5%)	11	(26.8%)	5	(31.3%)
Treatment, $N$ (%)									0.038							0.19
Surgical	42	(87.5%)	13	(86.7%)	11	(64.7%)	8	(14.3%)		33	(89.2%)	31	(75.6%)	10	(62.5%)
Chemotherapy/other	4	(8.33%)	0	(0%)	1	(5.9%)	1	(7.1%)		2	(5.4%)	2	(4.9%)	2	(12.5%)
Radiation therapy	2	(4.17%)	2	(13.3%)	5	(29.4%)	5	(35.7%)		2	(5.4%)	8	(19.5%)	4	(25.0%)
Staging, $N$ (%)									0.04							0.02
Early stage (I and II)	42	(87.5%)	14	(93.3%)	11	(64.7%)	9	(64.3%)		33	(89.2%)	34	(82.9%)	9	(56.3%)
Late stage (III and IV)	6	(12.5%)	1	(6.7%)	6	(35.3%)	5	(35.7%)		4	(10.8%)	7	(17.1%)	7	(43.8%)
Baseline screening, $N$ (%)									0.11							0.12
Positive (T0 $+$ )	26	(54.2%)	5	(33.3%)	4	(23.5%)	5	(35.7%)		20	(54.1%)	16	(39.0%)	4	(25.0%)
Negative (T0 $-$ )	22	(45.8%)	10	(66.7%)	13	(76.5%)	9	(64.3%)		17	(45.9%)	25	(61.0%)	12	(75.0%)

Abbreviations: sd $=$ standard deviation; FH $=$ family history; Pack-years $=$ packs smoked/day x years smoked; COPD $=$ chronic obstructive pulmonary disease; NOS $=$ not otherwise specified. ${}^{1}$ $P$ -value obtained from Chi-squared for categorical variables. ${}^{2}$ $P$ -value obtained from Anova for continuous variables. ${}^{3}$ $P$ -value obtained from Log-rank for survival variables. ${}^{4}$ BAC and adenocarcinoma were combined into one group.

Table 3

Multivariable Cox proportional hazards models for overall survival

	Screening interval from T0 to T1				Screening interval from T0 to T1 or T1 to T2
Characteristic	All patients ${}^{2}$ HR (95% CI)		Early-stage patients HR (95% CI)		All patients ${}^{2}$ HR (95% CI)		Early-stage patients HR (95% CI)
Risk group
Low	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)
Intermediate	1.12	(0.30, 4.13)	2.28	(0.37, 14.08)	2.38	(0.85, 6.61)	3.38	(0.85, 13.45)
High	3.12	(1.11, 8.75)	9.66	(1.37, 67.99)	3.62	(1.12, 11.69)	7.82	(1.44, 42.43)
Very high	10.90	(3.98,29.85)	21.36	(3.70, 123.60)	n/a		n/a
Sex
Female	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)
Male	1.61	(0.68, 3.85)	2.96	(0.55, 15.97)	1.46	(0.62, 3.43)	1.96	(0.60, 6.40)
Stage
I and II	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)
III and IV	8.25	(3.42, 19.90)	n/a		7.03	(3.10, 15.93)	n/a
Histology
Adenocarcinoma-BAC ${}^{1}$	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)
Squamous	2.84	(1.01. 7.97)	3.05	(0.73, 12.78)	1.38	(0.55, 3.47)	1.08	(0.35, 3.35)
Small cell	1.45	(0.33, 6.42)	n/a		1.52	(0.32, 7.23)	n/a
Other NOS	0.84	(0.28, 2.53)	0.31	(0.04, 2.11)	0.66	(0.26, 1.65)	0.69	(0.19, 2.53)
Self-reported COPD
No	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)	1.00	(Reference)
Yes	2.49	(0.91, 6.81)	2.45	(0.61, 9.77)	2.10	(0.76, 5.81)	1.07	(0.29, 3.99)
Treatment
Surgical	1.00	(Reference)	1.00	(Reference)	n/a		n/a
Chemotherapy/other	6.58	(1.48, 29.36)	372.88	(12.99, 10707.83)	n/a		n/a
Radiotherapy	3.61	(1.43, 9.11)	2.13	(0.53, 8.51)	n/a		n/a
Harrell’s C-index	0.85		0.84		0.80		0.72
Abbreviations: COPD $=$ chronic obstructive pulmonary disease; NOS $=$ not otherwise specified; 1BAC and adenocarcinoma were combined into one group. Bold values represent statistically significant.

According to Aerts et al. [3] the formula is:

$\displaystyle F_{\textit{morph.comp.2}}=36\pi\frac{V^{2}}{A^{3}}$

To identify a potential prognostic threshold, compactness was included into CART to identify the most informative cutpoint associated with OS (Fig. 4A). A VDT $<$ 279 and Compactness $>=$ 0.60 had similar prognosis capabilities with poor outcomes (Fig. 4B).

Figure 4.

Intratumoral feature Compactness and VDT have similar prognosis capabilities. (A) The tree structure of the classification and regression tree analysis (CART) which identified the survival cutpoint of compactness. (B) Overall Survival for VDT dichotomized at 279 days and Compactness dichotomized at 0.60. Abbreviations: RHR $=$ relative hazard ratio.

3.5 Radiogenomics analysis

To reduce false positives in the radiogenomics analysis, we utilized a very high $p$ -value threshold ( $P<$ 0.0001) to identify gene probesets associated with Compactness. From the radiogenomic analyses, MIR34A was found to be most significant gene associated ( $P=$ 2.01E-05) with the compactness radiomic features (Fig. 5). When Compactness was dichotomized at the median, mean VDT was higher in Compactness low vs Compactness high (mean $=$ 999.84 [3522.55] vs 656.60 [SD $=$ 2345.99] ( $P=$ 0.007).

Figure 5.

Association between radiomics and gene expression. MIR34A expression by dichotomizing Compactness at the median.

4. Discussion

Overdiagnosis is a serious adverse effect of cancer screening and early detection. A recent meta-analysis of randomized trials on overdiagnosis of lung cancer with LDCT [7] estimated that 49% of the LDCT screen-detected cancers may be overdiagnosed. Though the authors conclude, “there is great uncertainty about the degree of overdiagnosis in lung cancer screening due to the heterogeneity” of screening trials, there is substantial evidence [7, 11, 18, 35, 42] that overdiagnosis is a real world problem and robust biomarkers such as radiomics could have translational implications by risk stratifying screen-detected lung cancer. To that end, we utilized VDT and peritumoral and intratumoral radiomic features to identify high-risk screen-detected lung cancers associated with poor survival outcomes. Among patients diagnosed with lung cancer at the first screening interval (T1), very-high risk patients had extremely poor survival outcomes (HR $=$ 11.71; 21.4% 5-year OS) versus the low-risk group (HR $=$ 1.00; 82.1% 5-year OS). Among early-stage patients, the very-high risk group had extremely poor survival outcomes (HR $=$ 11.71; 21.4% 5-year OS) versus the low-risk group (HR $=$ 1.00; 82.1% 5-year OS). When combined patients diagnosed with lung cancer at the first (T1) and second (T2) screening intervals, patients in the high-risk group also exhibited extremely poor survival outcomes (HR $=$ 8.15; 25% 5-year OS) versus the low-risk group (HR $=$ 1.00; 83.3% 5-year OS; C-index $=$ 0.69). Among early-stage patients, the high-risk patients had poor survival outcomes (HR $=$ 9.07; 44.4% 5-year OS) versus the low-risk group (HR $=$ 1.00; 90.9% 5-year OS).

VDT has been largely utilized as a diagnostic biomarker to discriminate malignant and benign nodules [8, 31, 53]; however, the thresholds to discriminate between malignant and benign pulmonary nodules are inconsistent. Additionally, prior studies utilized VDT as a biomarker to differentiate between indolent and aggressive lung cancers diagnosed [5, 21, 36, 50, 51] and have also utilized VDT as a prognostic factor [32, 40, 49]. For example, a VDT of less than 400 days was found to be an independent risk factor for poor disease-free survival [32]. In a study based on chest x-rays, the 5-year survival rate of fast-growing nodules with a VDT of $<$ 113 days was associated with significantly lower than slow-growing nodules with a VDT $>$ 113 days [49]. Setojima et al. separated whole tumor VDT (WVDT) and solid-part tumor VDT (SVDT) and identified cut-offs 214 days for WVDT and less than 215 days for SVDT that were associated with poor recurrence-free survival and overall survival [40]. The decision tree in our analysis identified novel VDT thresholds (VDTs of 234 and 279) that exhibited independent capacity to discriminate between high-risk/aggressive tumors and low-risk/indolent tumors. Because a VDT between 20 and 400 days represents a high likelihood of malignancy, the cumulative evidence suggests that this broad range is capturing heterogeneous behaviors, including lower risk cancers such as indolent and slowing growing tumors.

While prior studies by group have utilized radiomic features only [37] and the combination of radiomics and an a priori VDT threshold [29] (i.e., $<$ 400 vs. $\geqslant$ 400) as predictors of lung cancer behavior, this is the first study to identify novel VDT thresholds in combination of radiomics to characterize the lethal potential of lung cancers diagnosed in lung cancer screening. Another novel aspect of this work is that we identified a single timepoint, baseline radiomic feature (compactness) that predicts VDT and identified potential biological underpinnings of VDT (Fig. 4). Because VDT is a delta-radiomic that requires serial images over time to calculate, a radiomic feature from a single timepoint that could act as a surrogate could be important because VDT could be calculated through a proxy. Interestingly, the compactness radiomic feature had similar prognostic capabilities as VDT (Fig. 4B). Compactness is a measure of how compact the shape of the tumor is relative to a sphere and, much like VDT, has been found to be a diagnostic feature that differentiates malignant from benign nodules [22, 26]. Prior work has also show that compactness is associated with survival [3]. Additionally, our radiogenomics analyses revealed that high compactness is associated with a high expression of the MIR34A gene. MicroRNA 34a (miR-34a) is a microRNA that is encoded by the MIR34A gene and the miR-34 family members are tumor suppressive miRNAs appear to have a crucial role in repressing tumor progression by involving in epithelial-mesenchymal transition (EMT) via EMT-transcription factors, p53, and other signally pathways [19, 20, 54]. In prior studies, high expression of MIR34A in plasma and tumor tissue was associated with prolonged overall survival and disease-free survival among non-small cell lung cancer patients [55]. Interestingly, Aherne et al. found that MIR34A was significantly increased in the adenoma and early-stage cancer groups compared to controls [4]. However, the role of MIR34A on nodule growth and behavior is currently unknown.

We acknowledge some limitations of this study. First, the sample size is modest because we utilized lung cancer cases with specific inclusion and exclusion criteria from the NLST. However, we utilized bootstrapping for internal validation which revealed the models were well calibrated. Hopefully other lung cancer screening trials, such as NELSON [11], LUSI [6], and MILD [33], will make their images and data public so that independent validation analyses can be performed. Additional research is also needed to validate the biological underpinnings of the gene identified in the radiogenomics analysis.

The results from our analyses revealed that radiomics combined with VDT can identify a vulnerable subset of screen-detected lung cancers that are associated with poor survival outcome suggesting such patients may need more aggressive treatment, such as adjuvant therapies, and more aggressive surveillance/follow-up. Further research is needed to validate these findings in other lung cancer screening trials and programs and to fully elucidate the biological underpinnings of these radiomic features and VDT.

Author contributions

Conception: J.P-M, R.J.G, M.B.S

Interpretation or analysis of data: J.P-M, H.L, W.M, I.T,T.K, S.A.E, M.B.S

Preparation of the manuscript: J.P-M, H.L, W.M, I.T, T.K, S.A.E, Y.B, R.J.G, M.B.S

Revision for important intellectual content: J.P-M, H.L, W.M, I.T, T.K, S.A.E, Y.B, R.J.G, M.B.S

Supervision: M.B.S and R.J.G

Supplementary data

The supplementary files are available to download from http://dx.doi.org/10.3233/CBM-210194.

sj-docx-1-cbm-10.3233_CBM-210194.docx - Supplemental material

Supplemental material, sj-docx-1-cbm-10.3233_CBM-210194.docx

Footnotes

Acknowledgments

Funding support from U54-CA163068, U54-CA1630 71, P20 CA202920, U01-CA143062, and U01-CA2004 64. None of the authors are affiliated with the National Cancer Institute. The authors thank the NCI for access to NCI’s data collected by the National Lung Screening Trial. The statements contained herein are solely those of the authors and do not represent or imply concurrence or endorsement by the NCI.

Conflict of interest

R.J.G is a consultant for HealthMyne Imaging Systems. J.P-M, H.L, W.M, I.T, T.K, S.A.E, Y.B, R.J.G, and M.B.S. declare no competing interests.

References

Aberle

D.R.

Adams

A.M.

Berg

C.D.

Black

W.C.

Clapp

J.D.

Fagerstrom

R.M.

Gareen

I.F.

Gatsonis

Marcus

P.M.

and Sicks

J.D.

, Reduced lung-cancer mortality with low-dose computed tomographic screening, N Engl J Med 365 (2011), 395–409.

Aberle

D.R.

Berg

C.D.

Black

W.C.

Church

T.R.

Fagerstrom

R.M.

Galen

Gareen

I.F.

Gatsonis

Goldin

Gohagan

J.K.

Hillman

Jaffe

Kramer

B.S.

Lynch

Marcus

P.M.

Schnall

Sullivan

D.C.

Sullivan

and Zylak

C.J.

, The National Lung Screening Trial: Overview and study design, Radiology 258 (2011), 243–253.

Aerts

H.J.W.L.

Velazquez

E.R.

Leijenaar

R.T.H.

Parmar

Grossmann

Carvalho

Bussink

Monshouwer

Haibe-Kains

Rietveld

Hoebers

Rietbergen

M.M.

Leemans

C.R.

Dekker

Quackenbush

Gillies

R.J.

and Lambin

, Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach, Nature Communications 5 (2014), 4006.

Aherne

S.T.

Madden

S.F.

Hughes

D.J.

Pardini

Naccarati

Levy

Vodicka

Neary

Dowling

and Clynes

, Circulating miRNAs miR-34a and miR-150 associated with colorectal cancer progression, BMC Cancer 15 (2015), 329.

Baldwin

D.R.

Duffy

S.W.

Wald

N.J.

Page

Hansell

D.M.

and Field

J.K.

, UK Lung Screen (UKLS) nodule management protocol: Modelling of a single screen randomised controlled trial of low-dose CT screening for lung cancer, Thorax 66 (2011), 308–313.

Becker

Motsch

Trotter

Heussel

C.P.

Dienemann

Schnabel

P.A.

Kauczor

H.U.

Maldonado

S.G.

Miller

A.B.

Kaaks

and Delorme

, Lung cancer mortality reduction by LDCT screening-Results from the randomized German LUSI trial, Int J Cancer 146 (2020), 1503–1513.

Brodersen

Voss

Martiny

Siersma

Barratt

and Heleno

, Overdiagnosis of lung cancer with low-dose computed tomography screening: Meta-analysis of the randomised clinical trials, Breathe (Sheff) 16 (2020), 200013.

Callister

M.E.

Baldwin

D.R.

Akram

A.R.

Barnard

Cane

Draffan

Franks

Gleeson

Graham

Malhotra

Prokop

Rodger

Subesinghe

Waller

and Woolhouse

, British Thoracic Society guidelines for the investigation and management of pulmonary nodules, Thorax 70(Suppl 2) (2015), ii1–ii54.

Cherezov

Hawkins

S.H.

Goldgof

D.B.

Hall

L.O.

Liu

Balagurunathan

Gillies

R.J.

and Schabath

M.B.

, Delta radiomic features improve prediction for lung cancer incidence: A nested case-control analysis of the National Lung Screening Trial, Cancer Med 7 (2018), 6340–6356.

10.

Clay

Rajagopalan

Karwoski

Maldonado

Peikert

and Bartholmai

, Computer Aided Nodule Analysis and Risk Yield (CANARY) characterization of adenocarcinoma: radiologic biopsy, risk stratification and future directions, Transl Lung Cancer Res 7 (2018), 313–326.

11.

de Koning

H.J.

van der Aalst

C.M.

de Jong

P.A.

Scholten

E.T.

Nackaerts

Heuvelmans

M.A.

Lammers

J.-W.J.

Weenink

Yousaf-Khan

and Horeweg

N.J.N.E.j.o.m.

, Reduced lung-cancer mortality with volume CT screening in a randomized trial, 382 (2020), 503–513.

12.

Dreno

Thompson

J.F.

Smithers

B.M.

Santinami

Jouary

Gutzmer

Levchenko

Rutkowski

Grob

J.J.

Korovin

Drucis

Grange

Machet

Hersey

Krajsova

Testori

Conry

Guillot

Kruit

W.H.J.

Demidov

Thompson

J.A.

Bondarenko

Jaroszek

Puig

Cinat

Hauschild

Goeman

J.J.

van Houwelingen

H.C.

Ulloa-Montoya

Callegaro

Dizier

Spiessens

Debois

Brichard

V.G.

Louahed

Therasse

Debruyne

and Kirkwood

J.M.

, MAGE-A3 immunotherapeutic as adjuvant therapy for patients with resected, MAGE-A3-positive, stage III melanoma (DERMA): A double-blind, randomised, placebo-controlled, phase 3 trial, Lancet Oncol 19 (2018), 916–929.

13.

Fave

Zhang

Yang

Mackin

Balter

Gomez

Followill

Jones

A.K.

Stingo

Liao

Mohan

and Court

, Delta-radiomics features for the prediction of patient outcomes in non-small cell lung cancer, Sci Rep 7 (2017), 588.

14.

Gierada

D.S.

Pinsky

Nath

Chiles

Duan

and Aberle

D.R.

, Projected outcomes using different nodule sizes to define a positive CT lung cancer screening examination, J Natl Cancer Inst 106 (2014).

15.

Gillies

R.J.

and Schabath

M.B.

, Radiomics improves cancer screening and early detection, Cancer Epidemiol Biomarkers Prev 29 (2020), 2556–2567.

16.

Harrell

, Regression Modeling Strategies version 6.1-1, 2021.

17.

Hawkins

Wang

Liu

Garcia

Stringfield

Krewer

Cherezov

Gatenby

R.A.

Balagurunathan

Goldgof

Schabath

M.B.

Hall

and Gillies

R.J.

, Predicting Malignant Nodules from Screening CT Scans, J Thorac Oncol 11 (2016), 2120–2128.

18.

Heleno

Siersma

and Brodersen

, Estimation of overdiagnosis of lung cancer in low-dose computed tomography screening: A secondary analysis of the danish lung cancer screening trial, JAMA Intern Med 178 (2018), 1420–1422.

19.

Hermeking

, The miR-34 family in cancer and apoptosis, Cell Death & Differentiation 17 (2010), 193–199.

20.

Hermeking

, MicroRNAs in the p53 network: micromanagement of tumour suppression, Nat Rev Cancer 12 (2012), 613–626.

21.

Heuvelmans

M.A.

Oudkerk

de Bock

G.H.

de Koning

H.J.

Xie

van Ooijen

P.M.

Greuter

M.J.

de Jong

P.A.

Groen

H.J.

and Vliegenthart

, Optimisation of volume-doubling time cutoff for fast-growing lung nodules in CT lung cancer screening reduces false-positive referrals, Eur Radiol 23 (2013), 1836–1845.

22.

Jun

Xia

Jiangdian

Min

Yali

and Jie

, Prediction of malignant and benign of lung tumor using a quantitative radiomic method, Annu Int Conf IEEE Eng Med Biol Soc 2016 (2016), 1272–1275.

23.

Kanashiki

Tomizawa

Yamaguchi

Kurishima

Hizawa

Ishikawa

Kagohashi

and Satoh

, Volume doubling time of lung cancers detected in a chest radiograph mass screening program: Comparison with CT screening, Oncol Lett 4 (2012), 513–516.

24.

J.P.

Berman

E.J.

Kaur

Babb

J.S.

Bomsztyk

Greenberg

A.K.

Naidich

D.P.

and Rusinek

, Pulmonary Nodules: Growth rate assessment in patients by using serial CT and three-dimensional volumetry, Radiology 262 (2012), 662–671.

25.

Lambin

Rios-Velazquez

Leijenaar

Carvalho

van Stiphout

R.G.

Granton

Zegers

C.M.

Gillies

Boellard

Dekker

and Aerts

H.J.

, Radiomics: Extracting more information from medical images using advanced feature analysis, Eur J Cancer 48 (2012), 441–446.

26.

Limkin

E.J.

Reuzé

Carré

Sun

Schernberg

Alexis

Deutsch

Ferté

and Robert

, The complexity of tumor shape, spiculatedness, correlates with tumor radiomic shape features, Scientific Reports 9 (2019), 4329.

27.

Lindell

R.M.

Hartman

T.E.

Swensen

S.J.

Jett

J.R.

Midthun

D.E.

Tazelaar

H.D.

and Mandrekar

J.N.

, Five-year lung cancer screening experience: CT appearance, growth rate, location, and histologic features of 61 lung cancers, Radiology 242 (2007), 555–562.

28.

Liu

Wang

McGettigan

M.J.

Balagurunathan

Garcia

A.L.

Thompson

Z.J.

Heine

J.J.

Gillies

R.J.

and Schabath

M.B.

, Radiologic features of small pulmonary nodules and lung cancer risk in the national lung screening trial: A nested case-control study, Radiology 286 (2017), 298–306.

29.

Balagurunathan

Abdalah

M.A.

Garcia

A.L.

Gillies

R.J.

and Schabath

M.B.

, Multi-window CT based Radiomic signatures in differentiating indolent versus aggressive lung cancers in the National Lung Screening Trial: A retrospective study, Cancer Imaging 19 (2019), 45.

30.

Maisonneuve

Bagnardi

Bellomi

Spaggiari

Pelosi

Rampinelli

Bertolotti

Rotmensz

Field

J.K.

Decensi

and Veronesi

, Lung cancer risk prediction to select smokers for screening CT – a model based on the Italian COSMOS trial, Cancer Prev Res (Phila) 4 (2011), 1778–1789.

31.

Nair

Devaraj

Callister

M.E.J.

and Baldwin

D.R.

, The Fleischner Society 2017 and British Thoracic Society 2015 guidelines for managing pulmonary nodules: Keep calm and carry on, Thorax 73 (2018), 806.

32.

Park

Lee

S.M.

Kim

Lee

J.G.

Choi

K.H.

and Seo

J.B.

, Volume doubling times of lung adenocarcinomas: Correlation with predominant histologic subtypes and prognosis, Radiology 295 (2020), 703–712.

33.

Pastorino

Silva

Sestini

Sabia

Boeri

Cantarutti

Sverzellati

Sozzi

Corrao

and Marchianò

ï¼Œ Prolonged lung cancer screening reduced 10-year mortality in the MILD trial: New confirmation of lung cancer screening efficacy, Ann Oncol 30 (2019), 1162–1169.

34.

Patz

E.F.

, Jr Pinsky

and Kramer

B.S.

, Estimating Overdiagnosis in Lung Cancer Screening – ReplyLetters, JAMA Internal Medicine 174 (2014), 1198–1199.

35.

Patz

E.F.

, Jr Pinsky

Gatsonis

Sicks

J.D.

Kramer

B.S.

Tammemägi

M.C.

Chiles

Black

W.C.

and Aberle

D.R.

, Overdiagnosis in low-dose computed tomography screening for lung cancer, JAMA Intern Med 174 (2014), 269–274.

36.

Pedersen

J.H.

Ashraf

Dirksen

Bach

Hansen

Toennesen

Thorsen

Brodersen

Skov

B.G.

Døssing

Mortensen

Richter

Clementsen

and Seersholm

, The Danish randomized lung cancer CT screening trial-overall design and results of the prevalence round, J Thorac Oncol 4 (2009), 608–614.

37.

Pérez-Morales

Tunali

Stringfield

Eschrich

S.A.

Balagurunathan

Gillies

R.J.

and Schabath

M.B.

, Peritumoral and intratumoral radiomic features predict survival outcomes among patients diagnosed in lung cancer screening, Sci Rep 10 (2020), 10528.

38.

Schabath

M.B.

Massion

P.P.

Thompson

Z.J.

Eschrich

S.A.

Balagurunathan

Goldof

Aberle

D.R.

and Gillies

R.J.

, Differences in Patient Outcomes of Prevalence, Interval, and Screen-Detected Lung Cancers in the CT Arm of the National Lung Screening Trial, PLoS One 11 (2016), e0159880.

39.

Schabath

M.B.

Welsh

E.A.

Fulp

W.J.

Chen

Teer

J.K.

Thompson

Z.J.

Engel

B.E.

Xie

Berglund

A.E.

Creelan

B.C.

Antonia

S.J.

Gray

J.E.

Eschrich

S.A.

Chen

D.T.

Cress

W.D.

Haura

E.B.

and Beg

A.A.

, Differential association of STK11 and TP53 with KRAS mutation-associated gene expression, proliferation and immune surveillance in lung adenocarcinoma, Oncogene 35 (2016), 3209–3216.

40.

Setojima

Shimada

Tanaka

Shigefuku

Makino

Maehara

Hagiwara

Masuno

Yamada

Kakihana

Kajiwara

Ohira

, and Ikeda

, Prognostic impact of solid-part tumour volume doubling time in patients with radiological part-solid or solid lung cancer, Eur J Cardiothorac Surg 57 (2020), 763–770.

41.

Sone

Nakayama

Honda

Tsushima

Haniuda

Takahashi

Suzuki

Yamanda

Kondo

Hanaoka

Takayama

Kubo

and Fushimi

, Long-term follow-up study of a population-based 1996–1998 mass screening programme for lung cancer using mobile low-dose spiral computed tomography, Lung Cancer 58 (2007), 329–341.

42.

Srivastava

Koay

E.J.

Borowsky

A.D.

De Marzo

A.M.

Ghosh

Wagner

P.D.

and Kramer

B.S.

, Cancer overdiagnosis: A biological challenge and clinical dilemma, Nat Rev Cancer 19 (2019), 349–358.

43.

Ten Haaf

and de Koning

H.J.

, Overdiagnosis in lung cancer screening: Why modelling is essential, J Epidemiol Community Health 69 (2015), 1035–1039.

44.

Ten Haaf

Jeon

Tammemägi

M.C.

Han

S.S.

Kong

C.Y.

Plevritis

S.K.

Feuer

E.J.

de Koning

H.J.

Steyerberg

E.W.

and Meza

, Risk prediction models for selection of lung cancer screening candidates: A retrospective validation study, PLoS Med 14 (2017), e1002277.

45.

Thawani

McLane

Beig

Ghose

Prasanna

Velcheti

and Madabhushi

, Radiomics and radiogenomics in lung cancer: A review for the clinician, Lung Cancer 115 (2018), 34–41.

46.

Treskova

Aumann

Golpon

Vogel-Claussen

Welte

and Kuhlmann

, Trade-off between benefits, harms and economic efficiency of low-dose CT lung cancer screening: A microsimulation analysis of nodule management strategies in a population-based setting, BMC medicine 15 (2017), 162–162.

47.

Tunali

Gillies

R.J.

and Schabath

M.B.

, Application of radiomics and artificial intelligence for lung cancer precision medicine, Cold Spring Harb Perspect Med (2021).

48.

Tunali

Hall

L.O.

Napel

Cherezov

Guvenis

Gillies

R.J.

and Schabath

M.B.

, Stability and reproducibility of computed tomography radiomic features extracted from peritumoral regions of lung cancer lesions, Med Phys 46 (2019), 5075–5085.

49.

Usuda

Saito

Sagawa

Sato

Kanma

Takahashi

Endo

Chen

Sakurada

and Fujimura

, Tumor doubling time and prognostic assessment of patients with primary lung cancer, Cancer 74 (1994), 2239–2244.

50.

van Klaveren

R.J.

Oudkerk

Prokop

Scholten

E.T.

Nackaerts

Vernhout

van Iersel

C.A.

van den Bergh

K.A.

van ’t Westeinde

van der Aalst

Thunnissen

D.M.

Wang

Zhao

Gietema

H.A.

de Hoop

B.J.

Groen

H.J.

de Bock

G.H.

van Ooijen

Weenink

Verschakelen

Lammers

J.W.

Timens

Willebrand

Vink

Mali

and de Koning

H.J.

, Management of lung nodules detected by volume CT scanning, N Engl J Med 361 (2009), 2221–2229.

51.

Veronesi

Maisonneuve

Bellomi

Rampinelli

Durli

Bertolotti

and Spaggiari

, Estimating overdiagnosis in low-dose computed tomography screening for lung cancer: A cohort study, Ann Intern Med 157 (2012), 776–784.

52.

Welch

H.G.

and Black

W.C.

, Overdiagnosis in cancer, J Natl Cancer Inst 102 (2010), 605–613.

53.

Yankelevitz

D.F.

and Henschke

C.I.

, Does 2-year stability imply that pulmonary nodules are benign? AJR Am J Roentgenol 168 (1997), 325–328.

54.

Zhang

Liao

and Tang

, MicroRNA-34 family: A potential tumor suppressor and therapeutic candidate in cancer, J Exp Clin Cancer Res 38 (2019), 53.

55.

Zhao

Cheng

Chen

Liu

and Zhang

, Circulating microRNA-34 family low expression correlates with poor prognosis in patients with non-small cell lung cancer, Journal of Thoracic Disease 9 (2017), 3735–3746.

56.

Zwanenburg

Vallières

Abdalah

M.A.

Aerts

H.J.W.L.

Andrearczyk

Apte

Ashrafinia

Bakas

Beukinga

R.J.

Boellaard

Bogowicz

Boldrini

Buvat

Cook

G.J.R.

Davatzikos

Depeursinge

Desseroit

M.-C.

Dinapoli

Dinh

C.V.

Echegaray

El Naqa

Fedorov

A.Y.

Gatta

Gillies

R.J.

Goh

Götz

Guckenberger

S.M.

Hatt

Isensee

Lambin

Leger

Leijenaar

R.T.H.

Lenkowicz

Lippert

Losnegård

Maier-Hein

K.H.

Morin

Müller

Napel

Nioche

Orlhac

Pati

Pfaehler

E.A.G.

Rahmim

Rao

A.U.K.

Scherer

Siddique

M.M.

Sijtsema

N.M.

Socarras Fernandez

Spezi

Steenbakkers

R.J.H.M.

Tanadini-Lang

Thorwarth

Troost

E.G.C.

Upadhaya

Valentini

van Dijk

L.V.

van Griethuysen

van Velden

F.H.P.

Whybra

Richter

and Löck

, The image biomarker standardization initiative: Standardized quantitative radiomics for high-throughput image-based phenotyping, Radiology 295 (2020), 328–338.

Volume doubling time and radiomic features predict tumor behavior of screen-detected lung cancers

Abstract

BACKGROUND:

METHODS:

RESULTS:

CONCLUSION:

Keywords

1. Introduction

2. Materials and methods

2.1 NLST study population

2.3 Radiogenomics analysis

2.4 Statistical analysis

3. Results

3.1 Patient characteristics

Table 1 Patient characteristics

3.3 Screening intervals from T0 to T1 and T1 to T2

3.4 Predicting VDT using T0 radiomics features

Table 2 Patient characteristics by risk group

Author contributions

Supplementary data

sj-docx-1-cbm-10.3233_CBM-210194.docx - Supplemental material

Footnotes

Acknowledgments

Conflict of interest

References

Table 1
Patient characteristics

Table 2
Patient characteristics by risk group