Semi-Quantitative Analysis for Determining the Optimal Threshold Value on CT to Measure the Solid Portion of Pulmonary Subsolid Nodules

Abstract

Purpose

This study aimed to investigate the optimal threshold value in Hounsfield units (HU) on CT to detect the solid components of pulmonary subsolid nodules using pathologic invasive foci as reference.

Materials and Methods

Thin-section non-enhanced chest CT scans of 25 patients with pathologically confirmed minimally invasive adenocarcinoma were retrospectively reviewed. On CT images, the solid portion was defined as the area with higher attenuation than various HU thresholds ranging from −600 to −100 HU in 50-HU intervals. The solid portion was measured as the largest diameter on axial images and as the maximum diameter on multiplanar reconstruction images. A linear mixed model was used to evaluate bias in each threshold by using the pathological size of invasive foci as reference.

Results

At a threshold of −400 HU, the biases were lowest between the largest/maximum diameter of the solid portion of subsolid nodule and the size of invasive foci of the pathological specimen, with 0.388 and −0.0176, respectively. They showed insignificant difference (p = 0.2682, p = 0.963, respectively) at a threshold of −400 HU.

Conclusion

For quantitative analysis, −400 HU may be the optimal threshold to define the solid portion of subsolid nodules as a surrogate marker of invasive foci.

INTRODUCTION

Subsolid nodules (SSNs), also known as ground-glass nodules, include nodules with hazy increased attenuation that do not obscure underlying bronchial structures or pulmonary vessels (1). The presence of a solid component in SSNs suggests its invasiveness and malignancy. Therefore, the presence and size of the solid portion are important for risk stratification and treatment planning (2,3).

In the 2017 Fleischner guidelines, when the size of persistent SSNs with a solid component is more than 6 mm, malignancy is highly suspected compared to when the size is larger than 6 mm without a solid portion (2). According to the Lung CT Screening Reporting and Data System, SSNs with a solid component > 6 mm or a growing solid component are classified into category 4, whereas pure SSNs are categorized into category 2 or 3 (4). In lung cancer staging by the American Joint Committee on Cancer (8th edition), the largest diameter of a solid component of ground-glass nodules is measured on the multiplanar reconstruction (MPR) image, and diameters of up to 1, 2, and 3 cm are cT1a, cT1b, and cT1c, respectively (5).

Defining and measuring the solid portion of SSNs are important. However, defining SSNs is a challenging task (6,7). Conventionally, it is based on the visual finding of individual assessment. Furthermore, the assessment of SSNs depends on the qualitative description from the individual radiologists with different CT scan protocols. Therefore, intra-observer and inter-observer variability among experienced chest radiologists cannot be avoided (8,9,10). Quantitative measurement of SSNs and its solid component may improve the accuracy and reduce inter-observer variability for differential diagnosis of SSNs. Few studies have evaluated the threshold attenuation for the solid portion indicating invasion foci on CT (11,12,13).

Despite these approaches, a quantitative threshold to determine the solid component of ground-glass nodules has not been established. Therefore, our study aimed to establish the optimal CT threshold value in Hounsfield units (HU) for detecting the solid components of SSNs.

MATERIALS AND METHODS

The study was approved by the Institutional Review Board of Gangnam Severance Hospital. Clinical data were reviewed from medical records (IRB No. 3-2016-0306). Given the retrospective nature of the study and the use of anonymized data, requirements for informed consent were waived.

PATIENTS

Twenty-five patients who underwent surgical resection of persistent SSNs and had pathologically confirmed minimally invasive adenocarcinoma (MIA) in Gangnam Severance Hospital were consecutively and retrospectively included. MIA was defined as lepidic-predominant adenocarcinoma measuring up to 3 cm with an invasive component measuring up to 0.5 cm and lacking lymphatic, vascular, alveolar space, or pleural invasion (5).

CT PROTOCOL

Standard CT images were acquired using two CT scanners: a 16-slice multidetector CT (MDCT) scanner (Somatom Sensation 16; Siemens Medical Solutions, Erlangen, Germany) and a 64-slice MDCT scanner (Somatom Sensation 64; Siemens Medical Solutions). Non-enhanced CT scan was performed at a tube potential of 120 kVp with automatic exposure control (100–200 mA). Axial images were reconstructed at a section width of 1 mm with 1-mm interval by filtered back projection technique and sharp convolution kernel.

CT IMAGE ANALYSIS

The solid components of SSN more than the cut off value were assumed to be direct surrogate markers to the invasive foci of pathologic specimen. The solid component of SSNs was defined as the area with high attenuation than a certain threshold in HU.

All pre-operative CT images of enrolled patients were analyzed with a commercial program (Aquarius iNtuition Ver. 4.4.12; TeraRecon, Inc., Foster City, CA, USA). The largest diameter of the solid portion is estimated on the axial image, and the maximum diameter of the solid portion is estimated on the MPR image in this program by setting a specific value and masking it in the lung window setting (W: 1400; L: −500) (Fig. 1). The diameter of lesions more than the threshold attenuation on each image was measured independently by two radiologists (C.P. and T.K), with 10 years and 25 years of experience in chest CT, respectively. Various thresholds ranging from −600 to −100 HU were applied in 50-HU intervals.

Fig. 1. Measurement of the largest diameter of the solid portion on axial images.

Lesions with higher than threshold attenuation are regarded as solid portions and marked with blue. The largest diameter is then measured by an individual. Threshold attenuation is arranged from the top left (-600 HU) to the bottom right (-100 HU). The largest diameter of the solid portion decreases as threshold attenuation increases.

HU = Hounsfield units

CATEGORICAL ASSESSMENT

Twenty-five pathologically proven MIA SSNs were categorized into radiologic adenocarcinoma in situ (AIS), MIA, and invasive adenocarcinoma (IA) based on criteria of the measured solid portion on CT at each HU threshold from −600 HU to −100 HU: 0 mm, 0 mm to 5 mm, and larger than 5 mm. Classification was performed in both axial and MPR images, respectively.

STATISTICAL ANALYSIS

All continuous variables were expressed as mean ± standard deviation. Categorical variables were summarized as frequencies or percentages. Data normality was evaluated using Shapiro-Wilk test. Bias between CT measurement of the solid portion and invasive foci on pathologic specimen was defined as the mean difference between the largest diameter of the solid component on the CT image and pathological invasive focus. It was evaluated with linear mixed model using Bonferroni’s method for considering repeated measurements from various thresholds. Inter-observer reproducibility in measuring the solid portions on axial and MPR images was evaluated with the intraclass correlation coefficient. A p value < 0.05 was considered statistically significant. For analysis, R (version 3.3.1.; R Foundation for Statistical Computing, Vienna, Austria) was used.

RESULTS

PATIENT AND LESION CHARACTERISTICS

The mean age of the 25 patients was 55.9 ± 12.7 years. Seven male patients (28%) were included in this study. The mean size of SSN was 9.2 ± 1.33 mm, and the mean attenuation was −539.1 ± 96 HU. The mean size of invasive foci, confirmed by pathology, was 3.56 ± 1.33 mm.

SEMI-QUANTITATIVE CT ANALYSIS ON AXIAL AND MULTIPLANAR RECONSTRUCTED IMAGES

The largest size of the solid portion on the axial image at each HU threshold from −600 HU to −100 HU in 50 HU intervals is shown in Table 1. At a threshold of −400 HU on the axial image, the bias was lowest between the measured solid components of SSNs and pathologically confirmed invasive foci (0.388), showing insignificant difference (p = 0.2682) (Fig. 2).

Table 1. Maximum Sizes of the Solid Portion of Subsolid Nodules on Axial and MPR ImagesMaximum Sizes of the Solid Portion of Subsolid Nodules on Axial and MPR Images.

Threshold (n = 25)		Mean Size (mm) ± Standard Deviation	Mean Difference (Bias)	p-Value
−100 HU
	Axial	0.26 ± 0.68	3.296	< 0.001
	MPR	0.31 ± 0.87	3.247	< 0.001
−150 HU
	Axial	0.50 ± 0.94	3.057	< 0.001
	MPR	0.55 ± 1.06	3.014	< 0.001
−200 HU
	Axial	0.89 ± 1.24	2.668	< 0.001
	MPR	0.93 ± 1.31	2.630	< 0.001
−250 HU
	Axial	1.38 ± 1.58	2.182	< 0.001
	MPR	1.39 ± 1.60	2.630	< 0.001
−300 HU
	Axial	1.88 ± 1.67	2.171	< 0.001
	MPR	2.00 ± 1.67	1.565	< 0.001
−350 HU
	Axial	2.47 ± 1.88	1.094	0.002
	MPR	2.76 ± 2.22	0.804	0.034
−400 HU
	Axial	3.17 ± 1.95	0.388	0.268
	MPR	3.58 ± 2.39	−0.018	0.963
−450 HU
	Axial	4.07 ± 2.07	−0.514	0.142
	MPR	4.52 ± 2.39	−0.962	0.011
−500 HU
	Axial	4.97 ± 2.17	−1.409	< 0.001
	MPR	5.44 ± 2.47	−1.882	< 0.001
−550 HU
	Axial	6.25 ± 2.22	−2.693	< 0.001
	MPR	6.67 ± 2.34	−3.109	< 0.001
−600 HU
	Axial	7.25 ± 2.50	−3.686	< 0.001
	MPR	7.75 ± 2.56	−4.186	< 0.001

The mean size of pathologically confirmed invasive foci is 3.56 ± 1.33 mm.

HU = Hounsfield units, MPR = multiplanar reconstruction

Fig. 2. Bar graph of bias between invasive foci and largest diameter of the solid portion on axial images along CT HU threshold with mean value and 95% confidence interval.

The dotted line represents the mean size of pathologically confirmed invasive foci (3.56 mm).

HU = Hounsfield units

The maximum size of the solid portion on the MPR image at each HU threshold from −600 HU to −100 HU in 50 HU intervals is shown in Table 1. Compared with the analysis on the axial image, the solid lesion component was larger on the MPR image. The measurements of solid components were not significantly different from the invasive component size on pathology at −100 HU threshold (p = 0.963). Similarly, at a threshold of −400 HU on the multiplanar reconstructed image, the bias was lowest (−0.0176) between the measured solid components of SSNs and pathologically confirmed invasive foci (Fig. 3). Inter-observer agreement in measuring the solid portion on axial and MPR images was excellent, with 0.985 and 0.961, respectively.

Fig. 3. Bar graph of bias between invasive foci and maximum diameter of the solid portion on multiplanar reconstruction images along CT HU threshold with mean value and 95% confidence interval.

The dotted line represents the mean size of pathologically confirmed invasive foci (3.56 mm).

HU = Hounsfield units

CATEGORICAL ASSESSMENT

Radiologic classification of the 25 SSNs is summarized in Table 2 and Fig. 4. On both axial and MPR images, 25 pathologically proven MIA SSNs were categorized into 20 AIS and 5 MIA at −100 HU threshold and 4 MIA and 21 IA at −600 HU threshold. At a threshold of −400 HU on the axial image, 19 (76%) SSNs were classified into MIA, which showed the highest accuracy.

Table 2. Classification of SSNs by Size of the Solid Portion on CT.

Threshold (n = 25)		AIS	MIA	IA
−100 HU
	Axial	20	5	0
	MPR	20	5	0
−150 HU
	Axial	18	7	0
	MPR	18	7	0
−200 HU
	Axial	14	11	0
	MPR	14	11	0
−250 HU
	Axial	11	14	0
	MPR	11	14	0
−300 HU
	Axial	7	18	0
	MPR	7	18	0
−350 HU
	Axial	6	17	2
	MPR	6	15	4
−400 HU
	Axial	2	19	4
	MPR	2	15	8
−450 HU
	Axial	2	13	10
	MPR	2	11	12
−500 HU
	Axial	0	13	12
	MPR	0	9	16
−550 HU
	Axial	0	7	18
	MPR	0	5	20
−600 HU
	Axial	0	4	21
	MPR	0	4	21

AIS = adenocarcinoma in situ, HU = Hounsfield units, IA = invasive adenocarcinoma, MIA = minimally invasive adenocarcinoma, MPR = multiplanar reconstruction, SSNs = subsolid nodules

Fig. 4. Stacked bar graph of 25 MIA SSNs for classification into radiologic AIS, MIA, and IA based on criteria for the measured solid portion on CT at each HU threshold.

A, B. Radiologic classification of the 25 SSNs, based on axial images (A) and multiplanar reconstruction images (B).

AIS = adenocarcinoma in situ, HU = Hounsfield units, IA = invasive adenocarcinoma, MIA = minimally invasive adenocarcinoma, SSNs = subsolid nodules

DISCUSSION

This study demonstrated that a threshold value which defines the solid component of MIA on the CT image of −400 HU showed the best performance. Both the largest diameter and the maximum diameter of the solid portion exhibited the lowest difference with the diameter of the pathologically confirmed solid portion on −400 HU, without significant difference. Bias was lower in the maximum diameter on the MPR image than in the largest diameter on the axial image. This finding may also be because the maximum diameter is measured in 3D dimension in pathology (14). However, −400 HU threshold on the axial image showed the highest accuracy for predicting MIA (76%). A threshold of −400 HU to −300 HU on both axial and MPR images showed relatively high accuracy for predicting MIA. As the threshold increased from −600 HU to −100 HU, the solid portion on CT reduced with increasing number of SSNs classified as AIS, indicating underdiagnosis. By contrast, as the threshold decreased from −100 HU to −600 HU, the solid portion on CT enlarged with increasing number of SSNs classified as IA, indicating overdiagnosis. Clinicians can use the lower HU threshold if they do not want to underdiagnosis the MIA as an AIS. Conversely, if clinicians do not want to overdiagnosis the MIA as an IA, they can use the relatively higher HU threshold.

In CT scan, the HU is proportional to the degree of X-ray attenuation (15), and attenuation of each pixel is related to tissue density (16). Normal alveolar structure and air space represent very low density, which is near −950–−850 HU. However, as a tumor grows, cellularity increases with high density and appears as high HU on CT, which is higher than the threshold, although it is difficult to differentiate with fibrosis, vasculatures, or the collapse of alveolar spaces (17,18). According to the 2015 WHO classification, invasive foci of MIA are defined as either any histologic subtype other than a lepidic pattern (such as acinar, papillary, micropapillary, or solid) or tumor cells infiltrating myofibroblastic stroma (14). Thus, we can assume that the density or cellularity of invasive foci in SSNs may be higher than tissue density or cellularity of approximately −400 HU empirically as a result of this study, even though not exactly −400 HU, suggesting higher cellularity or density than non-invasive tumor lesion with lepidic cellular growth or normal structures.

Some studies expended effort to set HU threshold for the solid lesions of SSNs (11,12,13) and tend to suggest a relatively high HU value as threshold (11,12). Matsuguma et al. (11) used −160 HU for the cut off value on the lung window to define the solid portion of SSNs. Similarly, Ko et al. (12) used −188 HU for the threshold and regarded the part of the nodule that showed higher attenuation as a solid part. The Fleischner guidelines recommended measurement of the solid component with narrow and/or mediastinal setting, which may indicate HU threshold near −150 HU considering the width and level of mediastinal setting (2,19). Meanwhile, Cohen et al. (13) suggested that segmentation of the solid component shows the best performance at a threshold of −350 HU, and measurements on mediastinal windows (window width, 400 HU; level, 30 HU) were significantly smaller than the actual sizes of the invasive components. In the same context, Lee et al. (20) compared two different thresholds, namely, −160 HU and −400 HU. They concluded high diagnostic accuracy in −400 HU for determining the presence of solid components of SSNs, whereas low sensitivity and negative predictive value in −160 HU. Similarly, we showed −400 HU as a threshold value for defining the solid component of SSNs. The high HU threshold is a conservative suggestion, which may overlook the actual solid portion; however, as in our suggestion with a lower threshold, it can be observed that it fits well.

Several recent technological approaches are available to analyze SSN characteristics by using CT textures, radiomics, or even artificial intelligence and deep learning (21,22,23,24). Chae et al. (21) reported that higher kurtosis is a significant differentiator of pre-invasive lesions from IA in SSNs. Son et al. (22) argued that the 75th percentile CT attenuation value and entropy can help distinguish them. Moreover, radiomics and deep learning have recently shown high performance in distinguishing IA along SSNs with an area under the curve more than 0.9 (23,24). However, these complex methods are difficult to apply in daily practice, and further validation is needed. Our −400 HU threshold proposal is easy to apply as the default setting in PACS like the mediastinum setting, with excellent inter-observer agreement and high specificity for invasiveness, and even higher accuracy for real pathologically confirmed size of invasive foci.

This study has a number of limitations. First, it was a single-center retrospective study, and bias could not be avoided. Second, the study design only included pathologically confirmed MIA. No data were available for pre-invasive lesions, such as atypical adenomatous hyperplasia or AIS, and IA, and this can be considered as selection bias. Third, the sample size was small. A future multi-center study with large sample size involving other pre-invasive lesions is warranted. Finally, direct pathologic comparison was not conducted between CT showing solid portion and that exhibiting invasive foci of tissue. Solid portions of SSNs have a tendency to represent invasive foci, but they do not directly correspond with invasive foci (25). However, by setting a threshold attenuation of −400 HU for the solid portion of SSNs like in our study, the maximum diameter of the solid portion of SSNs can be easily and rapidly acquired without statistical difference with pathologically confirmed solid portion and with great inter-observer agreement.

In conclusion, −400 HU might be the optimal threshold to define the solid component of SSNs with the lowest bias and excellent inter-observer agreement. This threshold value is potentially helpful as a surrogate marker of invasive foci and to differentiate part-solid nodules from pure ground-glass nodules for guidelines.