Skip to main content
Download PDF

Advances in intelligent diagnosis methods for pulmonary ground-glass opacity nodules

BioMedical Engineering OnLine volume 17, Article number: 20 (2018) Cite this article

Abstract

Pulmonary nodule is one of the important lesions of lung cancer, mainly divided into two categories of solid nodules and ground glass nodules. The improvement of diagnosis of lung cancer has significant clinical significance, which could be realized by machine learning techniques. At present, there have been a lot of researches focusing on solid nodules. But the research on ground glass nodules started late, and lacked research results. This paper summarizes the research progress of the method of intelligent diagnosis for pulmonary nodules since 2014. It is described in details from four aspects: nodular signs, data analysis methods, prediction models and system evaluation. This paper aims to provide the research material for researchers of the clinical diagnosis and intelligent analysis of lung cancer, and further improve the precision of pulmonary ground glass nodule diagnosis.

Background

Lung cancer is one of the major causes of death in cancer patients, accounting for about 27% of all cancer deaths [1]. Pulmonary nodules are one of the early symptoms of lung cancer, and the correct diagnosis results can effectively improve the survival rate of patients with early lung cancer. According to the content of solid ingredients, pulmonary nodules can be divided into solid nodules and ground glass nodules (GGNs). At present, GGNs are more likely to be malignant than solid nodules [2]. But the research on GGNs is not enough. Therefore, this article mainly introduces the new progress of the diagnosis in the benign and malignant pulmonary ground glass nodules.

Pulmonary ground glass nodules are often referred to as pulmonary ground glass opacity (GGO). It is defined as the increased attention of the lung parenchyma without obscuration of the pulmonary vascular markings on CT images [3]. In addition, it is also known as the subsolid nodule (SSN), which is a characteristic but non-specific imaging sign. According to the presence of solid ingredients in the nodules, the GGO is divided into two types: the mixed ground glass opacity (mGGO) and the pure ground glass opacity (pGGO) [4]. The mGGO is also known as part-solid GGN, which contains solid part and non-solid part. But pGGO does not contain solid ingredients. The classification of pulmonary nodules is shown in Table 1. Here are some clinical images to describe the classification of pulmonary nodules in Fig. 1.

Table 1 The classification of pulmonary nodules
Full size table
Fig. 1
figure 1

Some different pulmonary nodules. a Irregular solid nodules in left lower lobe. b Pure ground glass nodules in left upper lobe. c Pure ground glass nodules in right upper lobe. d Mixed ground glass nodules in right upper lobe. e Mixed ground glass nodules in right middle lobe. f Mixed ground glass nodules in right middle lobe

Full size image

At present, there are already common processes for the lung nodules detection, including image acquisition, pre-processing, pulmonary parenchyma segmentation, nodule detection, and false positive reduction [5]. This process is consistent with the flow of the GGNs detection in the computer-aided detection (CAD) system. In the study of the detection and diagnosis of GGNs, the difficulty is the difference of the main emphasis of the researchers, the properties of datasets, the complexity of the system and the performance of evaluation methods. Considering the diversity of the existing works, this paper presents a review on GGNs diagnosis since 2014 to provide the research material for researchers of the clinical diagnosis and intelligent analysis of lung cancer. It can be divided into four sections including nodular signs, data analysis methods, prediction models and system evaluation. In the first section, the common nodular signs are enumerated. Subsequently, data analysis methods are used to find predictors among the nodular signs to predict benign and malignant GGNs, which can provide a reliable basis for the establishment of the follow-up benign and malignant prediction models. Next, establish predictive model for the differential diagnosis of benign and malignant nodules. The subsystem architecture for the diagnosis about GGNs in the CAD system can then be developed, hence summarize system evaluation methods to evaluate these system performances in the final section. Each section will be explained in further detail below.

Nodular signs

Objective and accurate understanding of nodular signs has important significance for benign and malignant diagnosis of GGNs. Combining the current research progress, some common nodular signs of pulmonary ground glass nodules in CAD system are described below. These nodular signs are respectively: nodule size or lesion diameter, lobulation and spiculation, smoking history, lung cancer history, the proportion of solid components within the lesion, pleural indentation, air bronchogram, and vascular convergence sign.

Nodule size or lesion diameter

Cho et al. [6] concluded that larger nodules were independent predictors of malignant tumors by analyzing 356 cases of ground glass nodules from 324 patients in January 2009–October 2013. Ueda et al. [7] found that the changes of CT images were associated with pathological findings. All non-malignant lesions were non-enlarged tumors, and all increased tumors were diagnosed with adenocarcinoma. Studies of Xu et al. [8] have shown that older patients with a greater nodular diameter are more likely to suffer lung cancer, and more than 10 mm of subsolid nodules are statistically significantly associated with malignant tumors. It can be seen that the nodule size or lesion diameter has a very close relationship with the benign and the malignant degree of pulmonary ground glass nodules. The greater the nodules indicate, the more likely the nodule is to be malignant.

Lobulation and spiculation

When a portion of the pulmonary nodule surface is wavy or fan-shaped, the edge of the nodule is described as the lobulation. Similarly, when the edge of the pulmonary nodule extends to the lung parenchyma, the nodule is called the spiculation. Hu et al. [9] conducted a study to determine the imaging characteristics that contributed to the differential diagnosis of solitary GGNs. Through the research on pathological examination results of 112 cases of 112 patients with solitary GGNs after surgical resection, it’s found that benign and malignant nodules have significant differences in lobulation, spiculation and other aspects. Liu et al. [10] presented a systematic approach to analyze low resolution CT images of 172 patients. Excluding nodule size in dataset 2, four best features to predict the malignant indicators of the original nodules were found, including lobulation and spiculation. Zhao et al. [11] focused on spiculation and found that the pulmonary resolving nodules and malignant nodules have similar CT features. So the lobulation and spiculation are the important indicators of malignant tumors.

Smoking history

The use of low-dose CT lung cancer screening can reduce lung cancer mortality in high-risk smokers [12]. However, it is far from enough. Some researchers have begun to study the relation between smokers and lung cancer mortality [13, 14]. She et al. [13] identified preinvasive lesions and invasive pulmonary adenocarcinomas (IPAs) based on data from patients with solitary pulmonary pure GGNs that had been confirmed from January 2009 to September 2015. It was found that smoking status was one of the predictors of the invasive extent. The longer the smoking time and the greater the amount of smoking, the higher the likelihood of suffering from malignant tumors.

Lung cancer history

Tamura et al. [15] reviewed 63 cases of pure ground glass nodules to assess the relationship between clinical and imaging findings and pulmonary GGN progression, identifying risk factors that predict pGGO lesions. In the growth group, pGGO lesions were closely related to the high mean computed tomography (m-CT) values and lung cancer history, but not close to smoking habits and GGN shape. It can be seen that lung cancer history is one of the independent predictors of future changes in GGN lesions. In the histological subtypes, the effects of clustering on lung cancer families still exist after a variety of confounding factors such as socioeconomic status and smoking habits. Wille et al. [16] demonstrated that the age and history of lung cancer in the Danish Lung Cancer Screening Trial (DLCST) have a significant predictive effect on the lung cancer risk of solitary pulmonary nodules (SPNs).

The proportion of solid components within the lesion

The pulmonary nodules in the order of the malignant risk are: mGGO>pGGO>solid nodules. In [2] and [14], it demonstrated the GGN is more likely to be malignant than solid nodules. But the proportion of solid components within the lesion is also important. Choi et al. [17] concluded that the presence of part-solid nodules were significantly associated with malignant lesions. Among them, the proportion of solid ingredients in GGNs and lung cancer clinicopathological staging is associated. The solid component of advanced lung cancer is significantly more than early lung cancer [18]. In addition, the focal ground glass opacity (fGGO) is an important sign of lung cancer. Its incidence of stage I lung cancer is higher than benign lesions. Malignant rates of mGGO, pGGO and solid lesions were 75.0, 60.0 and 48.2%, respectively. Studies showed that the later stage of lung cancer, the lower the proportion of pGGO, and the higher the proportion of mGGO. If fGGO lesions contain solid ingredients, it is the highest possibility of malignant lesions. Followed by pGGO, the possibility of malignant lesions is relatively low [19].

Pleural indentation

Pleural indentation, it is the linear, curly or star-shaped shadow between the tumor and pleura, which is closely related to the benign and malignant diagnosis of GGNs. In [10], researchers predict the malignant degree of the pulmonary nodules by using the imaging features. The pleural indentation is one of the four best feature sets in the dataset 1 that contains the size measurement. In order to study the pathological findings of pulmonary pGGO lesions and to assess the likelihood of invasive malignancies, Ichinose et al. [20] analyzed 191 patients with GGO lesions (114 cases of pGGO and 77 cases of mGGO) from January 2008 to December 2010 in 160 patients. They found that invasive lung cancer accounted for 12% of pGGO lesions, most of which showed pleural indentation. In [13], univariate analysis or multivariate analysis showed that pleural indentation was a predictor of lung adenocarcinoma invasion.

Air bronchogram

Air bronchogram, refers to the lesion that we can see the translucent bronchial shadow in the lung tissue area. Dai et al. [21] analyzed the pure GGNs data from 71 patients with primary tumors from June 2010 to December 2013. It was found that air bronchogram is one of the potential factors of pure GGNs. And it helps to determine the benign and malignant synchronous pure GGNs before surgery. In analyzing the CT characteristics and pathologic classification of patients with pGGO in early lung adenocarcinoma, Jin et al. [22] concluded that there was a close relationship between air bronchogram and histological invasiveness. It can help to predict the invasive degree of pGGO in early lung adenocarcinoma.

Vascular convergence sign

In [9], Hu et al. confirmed that vascular convergence sign is an important indicator of malignant lesions according to pathological examinations of 112 patients with 112 cases of solitary GGNs after surgical resection. It is one of the risk factors for characterizing malignant tumors. In the literature [14], the author also said that in addition to lobulation and spiculation, the vascular convergence sign is also one of the signs of malignant pulmonary solid nodules in imageology.

Yip et al. [23] reviewed and reanalyzed the existing literature, and concluded that lung cancer characterization for GGNs was a slow process. With the progress of CT scanning, GGNs are more frequently detected in chest CT scans of lung cancer and other indications, suggesting that physicians can have more time to cure patients with non-solid nodules. Although CT-guided percutaneous needle aspiration biopsy is an effective way to diagnose subsolid nodules in pathologic diagnosis [24], with the rapid development of computer technology, imaging signs to assess nodular growth and the benign and malignant situation are very useful in recent years. Especially in further image analysis, the use of CT enhancement studies and positron emission tomography (PET) techniques are useful in identifying malignant latent factors of GGNs [25]. The list of nodular signs associated with the benign and malignant identification in the relevant literature is shown in Table 2. It’s easy to see that the most frequently used nodular signs are nodule size or lesion diameter, lobulation and speculation, smoking history, pleural indentation, and lung cancer history. The rational use of several nodular signs can have a good diagnostic value for early lung cancer.

Table 2 The list of nodular signs in the relevant literature
Full size table

Data analysis methods

Based on the previous study of nodular signs, many scholars use statistical methods to assess the importance of these nodular signs and find predictors to predict benign and malignant GGNs effectively. Meanwhile, using data statistical methods can provide a reliable basis for researching the follow-up benign and malignant prediction models and differentiating malignant from benign pulmonary nodules. In addition to basic comparative statistics including Student’s t test, Chi square test, Fisher’s exact test, there are relatively novel or special data analysis methods such as density histogram, nomogram, volume doubling time (VDT) and mass doubling time (MDT). The appropriate use of data statistical methods can help radiologists to determine and identify the benign and malignant pulmonary nodules and help to reduce the pressure of radiologists.

Kamiya et al. [26] used the kurtosis and skewness of the density histogram to evaluate the characteristics of the pulmonary nodules. They found that the method was effective in assessing the characteristics of nodules and predicting the benign and malignant nodules. By analyzing 93 cases of pulmonary nodules (72 cases of malignant nodules and 21 cases of benign nodules), they found that the peak of malignant lesions was greater than benign nodules. It can be seen that the kurtosis and skewness of the density histogram can help distinguish benign and malignant nodules. In [13], the authors established and validated a new nomogram that identifies IPAs from preinvasive lesions in solitary pure GGN patients. It was found that lesion size, lesion edge, lesion shape, mean CT value, pleural indentation and smoking status were significantly correlated with the degree of invasion in univariate analysis or multivariate analysis.

Song et al. [27] estimated the volume doubling time and mass doubling time of low-dose CT persistent pulmonary subsolid nodules in patients with no history of malignant tumors, 97 SSNs of 97 patients were divided into three groups: Group A, pure GGNs; Group B, part-solid GGNs, the solid ingredients ≤ 5 mm; Group C, part-solid GGNs, the solid ingredients > 5 mm. The VDTs and MDTs of Group A and Group B were significantly higher than those in Group C. In other words, the VDTs and MDTs of pure GGNs and part-solid GGNs with solid ingredients ≤ 5 mm were significantly higher than those in part-solid GGNs with solid ingredients > 5 mm. Scholten et al. [28] also calculated the mass doubling time of pulmonary nodules by measuring the maximum diameter, volume, and mass. He analyzed 264 pulmonary subsolid nodules of 234 participants. The results showed that although persistent SSNs according to pathologic analysis had a higher rate of malignancy, it rarely developed into clinically malignant lesions.

Das et al. [29] analyzed the data from 32 patients with 35 pulmonary nodules (excluding pGGO, containing only solid nodules and mGGO), assessed the feasibility of diffuse kurtosis imaging (DKI) in the human lungs and compared its diagnostic value with standard diffusion weighted imaging in the differential diagnosis of benign and malignant pulmonary nodules. The results showed that human pulmonary DKI is clinically feasible. In theory, the mean peak (MK) of DKI in malignant lesions is significantly higher than that in benign lesions, showing a higher non-Gaussian diffusion. Dhara et al. [30] used differential geometry-based techniques to calculate the spiculation, lobulation, and sphericity using the binary mask of the segmented nodules. They proved that the boundary roughness of pulmonary nodules is an important sign of its degree of malignancy.

In addition, there are some studies on data analysis methods which can also contribute to the prediction of benign and malignant pulmonary nodules. Dhara et al. [31] studied the use of support vector machines (SVM) in the benign and malignant identification of pulmonary nodules (including solid nodules, subsolid nodules and non-solid nodules) using shape-based, edge-based, texture-based features to characterize these pulmonary nodules, with 891 nodules to verify the proposed classification program is superior to competing with other technologies. Kaya et al. [32] also explored the contribution of nodular features in the prediction of malignant lesions and proposed a classification method based on a weighted rule method to predict the benign and malignant pulmonary nodules. The experimental results showed that the classification results of malignant prediction using nodular features can be improved. In the literature cited above, data analysis methods and related nodular signs are shown in Table 3.

Table 3 The list of the data analysis methods in the relevant literature
Full size table

Some common nodular signs, including margin, volume, mass, lobulation, sphericity and texture, can be seen from Table 3. It can obviously see that scholars use different data statistical methods due to the different selection and combination of nodular signs. One of limitations is the inability to compare these data analysis methods. Although specific matters need to analyse specifically, finding the most widely used and recommended data analysis method is a problem needs to be solved.

Prediction models

In order to identify the malignant lesions in GGNs better, some scholars, with the basis of nodular signs and data statistical methods, established the prediction models to assess the patient’s malignant rate. The results of the research on the benign and malignant prediction models of GGNs are summarized as follows.

Shinohara et al. [33] collected age, smoking status, lung cancer history, nodule size, location, and spiculation data from 241 patients using the Mayo [34] prediction model, which was based on the American College of Chest Physicians (ACCP), to identify the probability of malignancy (POM) for each patient. Significant differences were found between benign and malignant patients in terms of age, smoking history, nodule size, and spiculation. But another finding was that the predictive model proposed by the ACCP guideline was not satisfactory in the differential diagnosis of benign and malignant solitary pulmonary nodules.

Zheng et al. [35] proposed a modified model for preoperatively predicting malignancy of the solitary pulmonary nodules, as shown in Table 4. Two-thirds of the 846 patients from the Fujian Medical University Union Hospital were randomly selected as a derivation set, and the remaining one-third was used as a validation set. They divided lesions according to the proportion of GGO. The probability of malignancy is ex/(1 + ex), where e is the base of natural logarithms. This prediction model accurately identified the malignant lesions of solitary pulmonary nodules, especially with 50% or greater of the GGO, which is superior to the Mayo Clinic model.

Table 4 The list of a modified preoperatively prediction model
Full size table

In [16], Wille et al. evaluated the discriminant properties of the PanCan model by conducting the Danish Lung Cancer Screening Trial (DLCST) in 1152 pulmonary nodules of 718 patients. It was found that the PanCan risk prediction model [36] has a high risk of lung cancer identification with solitary pulmonary nodules. Risk prediction of lung cancer is mainly based on the size of pulmonary nodules. In addition, spiculation, age, and family history also have a significant predictive effect. Van Riel et al. [37] compared the performance of the PanCan model, Lung-RADS [38] and 1.2016 NCCN guidelines [39] when identifying benign and malignant pulmonary nodules and determined the effect of different diameter measurement methods on model performance. The results showed that the performance of the PanCan model in the differential diagnosis of benign and malignant pulmonary nodules was superior to Lung-RADS and 1.2016 NCCN guidelines. The 1.2016 NCCN guidelines were most sensitive to the definition of nodule size. But the definition of nodule size doesn’t significantly affect Lung-RADS and the PanCan model. Takahashi et al. [40] proposed a new classification criterion of lung adenocarcinoma according to the theory from The International Association for the Study of Lung Cancer, American Thoracic Society, and European Respiratory Society (IASLC/ATS/ERS). The new classification criterion was based on histology, clinic, molecular, radiologic, surgical and other multidisciplinary methods. They found tumor invasiveness of lung adenocarcinoma is an important prognostic factor of pathologic stage IA lung adenocarcinoma. And the GGO ratio (the GGO ratio (%) defined as follows: (1 − [maximum dimension of consolidation on the lung window setting/maximum dimension of the tumor on the lung window setting]) × 100) is one of the effective indicators of invasive lung adenocarcinoma.

Recently, many researchers have also applied end-to-end learning machines in medical image analysis field. Two kinds of end-to-end machine learning methods are massive-training artificial neural networks (MTANNs) and convolutional neural networks (CNNs). Shen et al. [41] used the CNNs to establish an end-to-end computing architecture and studied high-level suspiciousness specific features for lung nodule classification with the Multi-crop Convolutional Neural Network (MC-CNN), which was robust in the feature extraction and malignancy suspiciousness classification of pulmonary nodules. Due to the removal of nodule segmentation and hand-crafted feature (e.g., texture and shape compactness) engineering work, the proposed method can simplify conventional lung nodule malignancy suspiciousness classification, which can help researchers to assess the uncertainty of malignancy. Tajbakhsh et al. [42] compared two end-to-end machine learning methods in the detection of pulmonary nodules and the performance of benign and malignant pulmonary nodules. Experiments showed that the performance of MTANNs in the detection and identification of pulmonary nodules was higher than that of CNNs when using only limited training data. When using a larger training dataset, the performance gap became less evident even though the margin was still significant.

The prediction models with a better performance have a significant predictive effect on the lung cancer risk. They can help radiologists reduce the reading time and improve the diagnostic accuracy. The traditional prediction models like the PanCan model have a better performance in identifying benign and malignant pulmonary nodules, but not yet enough. Besides, these prediction models that have been mentioned consider GGNs, not aimed at them. Lack of pertinences is one of the bottlenecks.

System evaluation

Combined with the first three stages, the complete subsystem architecture for the diagnosis about GGNs in the CAD system can be designed or built. The researcher adopted different evaluation methods to obtain and identify the suitable subsystem according to the individual specific conditions. The most commonly used evaluation methods are the receiver operating characteristics (ROC) and area under the receiver operator characteristic curve (AUC), such as [10, 22, 26], and so on.

Both [9] and [43] use binary logistic regression analysis methods to compare the relationship between imaging features and histopathological classification of benign and malignant pulmonary nodules.

Cha et al. [44] said that there was no single effective method for differential diagnosis of pulmonary nodules. But the growth rate measurement using volumetry, evaluation of tumor vascularity on dynamic helical CT, dual-energy CT and MRI, and physiologic evaluation of PET/CT can be used to characterize nodules. Shin et al. [45] also used volumetry. They measured interval changes in nodule volume using CT nodule volumetry software. The study showed that in low-dose CT scans if solid subcentimeter nodules were stable during the initial 2-year follow-up period, they can be regarded as benign lesions. Subcentimeter GGNs are more likely to grow than solid nodules, so they require a longer follow-up period.

Kobayashi et al. [46] used Cox proportional hazards and logistic regression models for evaluation. Experiments showed that the smoking history and initial lesion diameter were closely related to the growth of GGNs. Yanagawa et al. [47] also used the automated computer program to analyze the volume of lung adenocarcinoma in stage I, with the Cox proportional hazards and logistic regression model analysis. Two volume measurement methods (solid volume, ≥ 1.5 cm3; solid volume ratios, ≥ 63%) were found as independent predictors which was associated with an increased possibility of recurrence and/or death in patients with stage I adenocarcinoma.

There are a number of other evaluation methods. For example, Zhao et al. used means of generalized estimating equation analysis to compare CT features of resolving nodules with nonresolving nodules (stable and malignant) [11]. Gomez Saez et al. [48] used the Poisson regression method to calculate the risk and mortality of lung cancer in patients with solitary pulmonary nodules and found that nodule size, spiculation, and other imaging features were associated with lung cancer. Han et al. [49] quantitatively compared the characteristics of pulmonary nodules in CT images to reflect the powerful characteristics of malignant tumors. Unsworth [50] used three-dimensional morphological changes and edge sharpness analysis in the three-dimensional CT scan of malignant pulmonary nodules to complete false-positive reduction and malignancy classification. Jiang et al. [51] proposed a modified inflammation-based score and validated its effect on the malignant prediction of fGGO in the lungs.

Some major evaluation methods in all references are shown in Table 5. It is easy to see that the most commonly used evaluation methods are ROC, AUC, univariate analysis or multivariate analysis. These evaluation methods have stronger applicability and generality. Other methods use less due to the particularity of the application. It should not be supposed that the most used method is always better. It still depends on specific cases. In addition, none of research methods is limited to one evaluation method such as [16] and [22]. Everyone can choose the evaluation methods which are right for specific needs.

Table 5 The list of evaluation methods in all references
Full size table

In the literature cited above, the complete table about nodular signs, data analysis methods, prediction models, system evaluation and other information are shown in Table 6. As seen form the table, some studies can make horizontal comparison. For example, [16] and [37] all use same nodular signs (e.g., nodule size or lesion diameter, lobulation and spiculation), data analysis methods (e.g., Student’s t test), prediction models (e.g., PanCan model), system evaluation (AUC, ROC), data sets (e.g., DLCST database), and nodule style (e.g., pGGO, mGGO, and solid nodules). It is easy to further develop and improve in the GGN diagnosis. But most studies have different main emphasis, data sets, or nodule styles. Therefore, it’s difficult to systematically evaluate the accuracy, robustness and generalization of these studies. In addition, Table 6 indicates these models and methods considered the diagnosis of GGNs, but not aimed at them. As has been said above, GGNs are more likely to be malignant than solid nodules. The models or methods aiming at GGNs are more targeted and accurate. How to develop the models and methods aiming at GGNs is an outstanding issue. In short, further studies will be needed to develop the GGN diagnosis and formulate a unified reference standard to assess their performance.

Table 6 The complete table about nodular signs, data analysis methods, prediction models, system evaluation and other information
Full size table

Conclusions

In summary, in the study of benign and malignant diagnosis for pulmonary ground glass nodules, some basic factors or characteristics are more commonly used, such as nodule size, lobulation, spiculation, the proportion of solid components within the lesion, lung cancer history, pleural indentation and so on in nodular signs, density histogram, volume doubling time and mass doubling time in data analysis methods, and the Mayo, the PanCan and other prediction models. Based on these common features, with the combination of other targeted factors, the lung CAD system can effectively improve the detection efficiency of benign and malignant identification in ground glass nodules. However, the current prediction models and data analysis methods for ground glass nodules are still limited, and the existing prediction methods still have great potential for development. In addition, with the increasing importance of pulmonary ground glass nodules diagnosis in the early diagnosis of lung cancer, methods such as deep learning and other artificial intelligence methods also significantly improve the performance of the Lung CAD system. This paper lists the relevant studies in recent years, it’s expected that with this paper the technical and clinical researchers could work out new prediction methods with better accuracy and better convenience.

Abbreviations

GGNs:

ground glass nodules

GGO:

ground glass opacity

CT:

computed tomography

SSN:

subsolid nodule

mGGO:

mixed ground glass opacity

pGGO:

pure ground glass opacity

CAD:

computer-aided detection

IPAs:

invasive pulmonary adenocarcinomas

m-CT:

mean computed tomography

DLCST:

Danish Lung Cancer Screening Trial

SPNs:

solitary pulmonary nodules

fGGO:

the focal ground glass opacity

PET:

positron emission tomography

VDT:

volume doubling time

MDT:

mass doubling time

DKI:

diffuse kurtosis imaging

MK:

mean peak

SVM:

support vector machines

LU-RADS:

Lung-Reporting and Data System

ACCP:

American College of Chest Physicians

POM:

probability of malignancy

IASLC:

The International Association for the Study of Lung Cancer

ATS:

American Thoracic Society

ERS:

European Respiratory Society

MTANNs:

massive-training artificial neural networks

CNNs:

convolutional neural networks

MC-CNN:

Multi-crop Convolutional Neural Network

ROC:

receiver operating characteristics

AUC:

area under the receiver operator characteristic curve

CBIR:

content-based image retrieval

NCI:

National Cancer Institute

LIDC:

Lung Image Database Consortium

SDAE:

stacked denoising auto-encoder

References

  1. Siegel R, Jemal A. Cancer facts & figures 2015. American Cancer Society Cancer Facts & Figures 2015; Atlanta: American Cancer Society; 2015.

  2. Yaguchi A, Okazaki T, Takeguchi T, Matsumoto S, Ohno Y, Aoyagi K, Yamagata H. Semi-automated segmentation of solid and GGO nodules in lung CT images using vessel-likelihood derived from local foreground structure. In: Hadjiiski LM, Tourassi GD, editors. Proceedings of SPIE medical imaging 2015: computer-aided diagnosis, vol. 9414. 2015.

  3. Miller WT, Shah RM. Isolated diffuse ground-glass opacity in thoracic CT: causes and clinical presentations. Am J Roentgenol. 2005;184(2):613–22.

    Article  Google Scholar 

  4. Nakata M, Saeki H, Takata I, Segawa Y, Mogami H, Mandai K, Eguchi K. Focal ground-glass opacity detected by low-dose helical CT. Chest. 2002;121(5):1464–7.

    Article  Google Scholar 

  5. Lee SLA, Kouzani AZ, Hu EJ. Automated detection of lung nodules in computed tomography images: a review. Mach Vis Appl. 2012;23(1):151–63.

    Article  Google Scholar 

  6. Cho J, Ko S-J, Kim SJ, Lee YJ, Park JS, Cho Y-J, Yoon HI, Cho S, Kim K, Jheon S, et al. Surgical resection of nodular ground-glass opacities without percutaneous needle aspiration or biopsy. BMC Cancer. 2014;14:838.

    Article  Google Scholar 

  7. Ueda K, Hayashi M, Tanaka N, Hoshii Y, Tanaka T, Hamano K. Surgery for undiagnosed ground glass pulmonary nodules: decision making using serial computed tomography. World J Surg. 2015;39(6):1452–9.

    Article  Google Scholar 

  8. Xu X, Yao Y, Shen Y, Zhang P, Zhou J. Clinical analysis of percutaneous computed tomography-guided hook wire localization of 168 small pulmonary nodules. Ann Thorac Surg. 2015;100(5):1861–7.

    Article  Google Scholar 

  9. Hu H, Wang Q, Tang H, Xiong L, Lin Q. Multi-slice computed tomography characteristics of solitary pulmonary ground-glass nodules: differences between malignant and benign. Thorac Cancer. 2016;7(1):80–7.

    Article  Google Scholar 

  10. Liu Y, Balagurunathan Y, Atwater T, Antic S, Li Q, Walker RC, Smith GT, Massion PP, Schabath MB, Gillies RJ. Radiological image traits predictive of cancer status in pulmonary nodules. Clin Cancer Res. 2017;23(6):1442–9.

    Article  Google Scholar 

  11. Zhao YR, Heuvelmans MA, Dorrius MD, van Ooijen PMA, Wang Y, de Bock GH, Oudkerk M, Vliegenthart R. Features of resolving and nonresolving indeterminate pulmonary nodules at follow-up CT: the NELSON study. Radiology. 2014;270(3):872–9.

    Article  Google Scholar 

  12. Chen C-Y, Chen C-H, Shen T-C, Cheng W-C, Hsu C-N, Liao C-H, Chen C-Y, Hsia T-C, Liao W-C, Tu C-Y, et al. Lung cancer screening with low-dose computed tomography: experiences from a tertiary hospital in Taiwan. J Formos Med Assoc. 2016;115(3):163–70.

    Article  Google Scholar 

  13. She Y, Zhao L, Dai C, Ren Y, Zha J, Xie H, Jiang S, Shi J, Shi S, Shi W, et al. Preoperative nomogram for identifying invasive pulmonary adenocarcinoma in patients with pure ground-glass nodule: a multi-institutional study. Oncotarget. 2017;8(10):17229–38.

    Article  Google Scholar 

  14. Wang T, Yan T, Wan F, Ma S, Wang K, Wang J, Song J, He W, Bai J, Jin L. Surgical treatment of small pulmonary nodules under video-assisted thoracoscopy (a report of 129 cases). Zhongguo fei ai za zhi. 2017;20(1):35–40.

    Google Scholar 

  15. Tamura M, Shimizu Y, Yamamoto T, Yoshikawa J, Hashizume Y. Predictive value of one-dimensional mean computed tomography value of ground-glass opacity on high-resolution images for the possibility of future change. J Thorac Oncol. 2014;9(4):469–72.

    Article  Google Scholar 

  16. Wille MMW, van Riel SJ, Saghir Z, Dirksen A, Pedersen JH, Jacobs C, Thomsen LH, Scholten ET, Skovgaard LT, van Ginneken B. Predictive accuracy of the PanCan lung cancer risk prediction model -external validation based on CT from the Danish Lung Cancer Screening Trial. Eur Radiol. 2015;25(10):3093–9.

    Article  Google Scholar 

  17. Choi SM, Heo EY, Lee J, Park YS, Lee C-H, Park CM, Kang CH, Yim J-J, Kim YT, Yoo C-G, et al. Characteristics of benign solitary pulmonary nodules confirmed by diagnostic video-assisted thoracoscopic surgery. Clin Respir J. 2016;10(2):181–8.

    Article  Google Scholar 

  18. Takashima S, Maruyama Y, Hasegawa M, Yamanda T, Honda T, Kadoya M, Sone S. CT findings and progression of small peripheral lung neoplasms having a replacement growth pattern. Am J Roentgenol. 2003;180(3):817–26.

    Article  Google Scholar 

  19. Huang Y, Wang Y-J, Wang W-Y, Pu Q, Li W-M. Value of fGGO in diagnosing stage I lung cancers. Sichuan da xue xue bao Yi xue. 2014;45(2):316–9.

    Google Scholar 

  20. Ichinose J, Kohno T, Fujimori S, Harano T, Suzuki S, Fujii T. Invasiveness and malignant potential of pulmonary lesions presenting as pure ground-glass opacities. Ann Thorac Cardiovas Surg. 2014;20(5):347–52.

    Article  Google Scholar 

  21. Dai C, Ren Y, Xie H, Jiang S, Fei K, Jiang G, Chen C. Clinical and radiological features of synchronous pure ground-glass nodules observed along with operable non-small cell lung cancer. J Surg Oncol. 2016;113(7):738–44.

    Article  Google Scholar 

  22. Jin X, Zhao S-H, Gao J, Wang D-J, Wu J, Wu C-C, Chang R-P, Ju H-Y. CT characteristics and pathological implications of early stage (T1N0M0) lung adenocarcinoma with pure ground-glass opacity. Eur Radiol. 2015;25(9):2532–40.

    Article  Google Scholar 

  23. Yip R, Wolf A, Tam K, Taioli E, Olkin I, Flores RM, Yankelevitz DF, Henschke CI. Outcomes of lung cancers manifesting as nonsolid nodules. Lung Cancer. 2016;97:35–42.

    Article  Google Scholar 

  24. Munir S, Koppikar S, Hopman WM, Boag AH, Dhillon G, Salahudeen SR, Nolan RL, Flood J. Diagnostic yield for cancer and diagnostic accuracy of computed tomography-guided core needle biopsy of subsolid pulmonary lesions. J Thorac Imaging. 2017;32(1):50–6.

    Article  Google Scholar 

  25. Truong MT, Ko JP, Rossi SE, Rossi I, Viswanathan C, Bruzzi JF, Marom EM, Erasmus JJ. Update in the evaluation of the solitary pulmonary nodule. Radiographics. 2014;34(6):1658–79.

    Article  Google Scholar 

  26. Kamiya A, Murayama S, Kamiya H, Yamashiro T, Oshiro Y, Tanaka N. Kurtosis and skewness assessments of solid lung nodule density histograms: differentiating malignant from benign nodules on CT. Jpn J Radiol. 2014;32(1):14–21.

    Article  Google Scholar 

  27. Song YS, Park CM, Park SJ, Lee SM, Jeon YK, Goo JM. Volume and mass doubling times of persistent pulmonary subsolid nodules detected in patients without known malignancy. Radiology. 2014;273(1):276–84.

    Article  Google Scholar 

  28. Scholten ET, de Jong PA, de Hoop B, van Klaveren R, van Amelsvoort-van de Vorst S, Oudkerk M, Vliegenthart R, de Koning HJ, van der Aalst CM, Vernhout RM, et al. Towards a close computed tomography monitoring approach for screen detected subsolid pulmonary nodules? Eur Respir J. 2015;45(3):765–73.

    Article  Google Scholar 

  29. Das SK, Yang DJ, Wang JL, Zhang C, Yang HF. Non-Gaussian diffusion imaging for malignant and benign pulmonary nodule differentiation: a preliminary study. Acta Radiol. 2017;58(1):19–26.

    Article  Google Scholar 

  30. Dhara AK, Mukhopadhyay S, Saha P, Garg M, Khandelwal N. Differential geometry-based techniques for characterization of boundary roughness of pulmonary nodules in CT images. Int J Comput Assist Radiol Surg. 2016;11(3):337–49.

    Article  Google Scholar 

  31. Dhara AK, Mukhopadhyay S, Dutta A, Garg M, Khandelwal N. A combination of shape and texture features for classification of pulmonary nodules in lung CT images. J Digit Imaging. 2016;29(4):466–75.

    Article  Google Scholar 

  32. Kaya A, Can AB. A weighted rule based method for predicting malignancy of pulmonary nodules by nodule characteristics. J Biomed Inform. 2015;56:69–79.

    Article  Google Scholar 

  33. Shinohara S, Hanagiri T, Takenaka M, Chikaishi Y, Oka S, Shimokawa H, Nakagawa M, Uramoto H, So T, Aoki T, Tanaka F. Evaluation of undiagnosed solitary lung nodules according to the probability of malignancy in the American College of Chest Physicians (ACCP) evidence-based clinical practice guidelines. Radiol Oncol. 2014;48(1):50–5.

    Article  Google Scholar 

  34. Swensen SJ, Silverstein MD, Ilstrup DM, Schleck CD, Edell ES. The probability of malignancy in solitary pulmonary nodules—application to small radiologically indeterminate nodules. Arch Intern Med. 1997;157(8):849–55.

    Article  Google Scholar 

  35. Zheng B, Zhou X, Chen J, Zheng W, Duan Q, Chen C. A modified model for preoperatively predicting malignancy of solitary pulmonary nodules: an Asia cohort study. Ann Thorac Surg. 2015;100(1):288–94.

    Article  Google Scholar 

  36. McWilliams A, Tammemagi MC, Mayo JR, Roberts H, Liu G, Soghrati K, Yasufuku K, Martel S, Laberge F, Gingras M, et al. Probability of cancer in pulmonary nodules detected on first screening CT. N Engl J Med. 2013;369(10):910–9.

    Article  Google Scholar 

  37. van Riel SJ, Ciompi F, Jacobs C, Winkler Wille MM, Scholten ET, Naqibullah M, Lam S, Prokop M, Schaefer-Prokop C, van Ginneken B. Malignancy risk estimation of screen-detected nodules at baseline CT: comparison of the PanCan model, Lung-RADS and NCCN guidelines. Eur Radiol. 2017;27:4019–29.

    Article  Google Scholar 

  38. Radiology ACo: Lung CT screening reporting and data system (Lung-RADS). 2014. http://www.acrorg/Quality-Safety/Resources/LungRADS. p. 11.

  39. Network NCC: Lung Cancer Screening (version 1. 2016). 2016. Also available at: http://www.nccn.org/professionals/physician_gls/f_guidelines.asp#detection. Accessed 23 June 2015.

  40. Takahashi M, Shigematsu Y, Ohta M, Tokumasu H, Matsukura T, Hirai T. Tumor invasiveness as defined by the newly proposed IASLC/ATS/ERS classification has prognostic significance for pathologic stage IA lung adenocarcinoma and can be predicted by radiologic parameters. J Thorac Cardiovasc Surg. 2014;147(1):54–9.

    Article  Google Scholar 

  41. Shen W, Zhou M, Yang F, Yu D, Dong D, Yang C, Zang Y, Tian J. Multi-crop Convolutional Neural Networks for lung nodule malignancy suspiciousness classification. Pattern Recogn. 2017;61:663–73.

    Article  Google Scholar 

  42. Tajbakhsh N, Suzuki K. Comparing two classes of end-to-end machine-learning models in lung nodule detection and classification: MTANNs vs. CNNs. Pattern Recogn. 2017;63:476–86.

    Article  Google Scholar 

  43. Liu L-H, Liu M, Wei R, Jin E-H, Liu Y-H, Xu L, Li W-W, Huang Y. CT findings of persistent pure ground glass opacity: can we predict the invasiveness? Asian Pac J Cancer Prev APJCP. 2015;16(5):1925–8.

    Article  Google Scholar 

  44. Cha MJ, Lee KS, Kim HS, Lee SW, Jeong CJ, Kim EY, Lee HY. Improvement in imaging diagnosis technique and modalities for solitary pulmonary nodules: from ground-glass opacity nodules to part-solid and solid nodules. Expert Rev Respir Med. 2016;10(3):261–78.

    Article  Google Scholar 

  45. Shin KE, Lee KS, Yi CA, Chung MJ, Shin M-H, Choi Y-H. Subcentimeter lung nodules stable for 2 years at LDCT: long-term follow-up using volumetry. Respirology. 2014;19(6):921–8.

    Article  Google Scholar 

  46. Kobayashi Y, Sakao Y, Deshpande GA, Fukui T, Mizuno T, Kuroda H, Sakakura N, Usami N, Yatabe Y, Mitsudomi T. The association between baseline clinical-radiological characteristics and growth of pulmonary nodules with ground-glass opacity. Lung Cancer. 2014;83(1):61–6.

    Article  Google Scholar 

  47. Yanagawa M, Tanaka Y, Leung AN, Morii E, Kusumoto M, Watanabe S, Watanabe H, Inoue M, Okumura M, Gyobu T, et al. Prognostic importance of volumetric measurements in stage I lung adenocarcinoma. Radiology. 2014;272(2):557–67.

    Article  Google Scholar 

  48. Gomez-Saez N, Hernandez-Aguado I, Vilar J, Gonzalez-Alvarez I, Fermina Lorente M, Luisa Domingo M, Pastor Valero M, Anne Parker L, Lumbreras B. Lung cancer risk and cancer-specific mortality in subjects undergoing routine imaging test when stratified with and without identified lung nodule on imaging study. Eur Radiol. 2015;25(12):3518–27.

    Article  Google Scholar 

  49. Han F, Wang H, Zhang G, Han H, Song B, Li L, Moore W, Lu H, Zhao H, Liang Z. Texture feature analysis for computer-aided diagnosis on pulmonary nodules. J Digit Imaging. 2015;28(1):99–115.

    Article  Google Scholar 

  50. Unsworth A. Proceedings of the Institution of Mechanical Engineers, Part H: Journal of Engineering in Medicine. 2016.

  51. Jiang L, Jiang S, Lin Y, Yang H, Zhao Z, Xie Z, Lin Y, Long H. Modified inflammation-based score as an independent malignant predictor in patients with pulmonary focal ground-glass opacity: a propensity score matching analysis. Sci Rep. 2016;6:19105.

    Article  Google Scholar 

Download references

Authors’ contributions

JY and HW conceived and designed the manuscript. CG, YD and JJ critically revised the manuscript. All authors read and approved the final manuscript.

Acknowledgements

Not applicable.

Competing interests

The authors declare that they have no competing interests.

Availability of data and materials

Not applicable.

Consent for publication

Not applicable.

Ethics approval and consent to participate

Not applicable.

Funding

This work was supported by Zhejiang Basic Public Welfare Research in 2018 (LGF18H160035), National Key Research & Development Plan (2017YFB1103602, 2017YFB1103600), NSFC Grants (61501452), Jiangsu Key Research and Development programs (BE2016613, BE2016010, BE2016010-3, BE2016010-4,BE2017675), Zhejiang Key Technology RD Program (2013C03010), Suzhou Industry Technological Innovation Projects (SYG201606, SYG201706), SND Medical Plan Project (2016Z010), and National 863 Program of China (2015AA020514).

Publisher’s Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Author information

Author notes

  1. Jing Yang and Hailin Wang contributed equally to this work and should be considered co-first author

Authors and Affiliations

  1. School of Biomedical Engineering, University of Science and Technology of China, Hefei, 230026, People’s Republic of China

    Jing Yang

  2. Suzhou Institute of Biomedical Engineering and Technology, Chinese Academy of Sciences, Suzhou, 215163, People’s Republic of China

    Jing Yang, Chen Geng & Yakang Dai

  3. Radiology Department and Interventional Radiology Center, The Fifth Affiliated Hospital of Wenzhou Medical University, Affiliated Lishui Hospital of Zhejiang University, The Central Hospital of Zhejiang Lishui, Lishui, 323000, People’s Republic of China

    Hailin Wang & Jiansong Ji

Authors
  1. Jing Yang
    View author publications

    You can also search for this author in PubMed Google Scholar

  2. Hailin Wang
    View author publications

    You can also search for this author in PubMed Google Scholar

  3. Chen Geng
    View author publications

    You can also search for this author in PubMed Google Scholar

  4. Yakang Dai
    View author publications

    You can also search for this author in PubMed Google Scholar

  5. Jiansong Ji
    View author publications

    You can also search for this author in PubMed Google Scholar

Corresponding author

Correspondence to Jiansong Ji.

Rights and permissions

Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. The Creative Commons Public Domain Dedication waiver (http://creativecommons.org/publicdomain/zero/1.0/) applies to the data made available in this article, unless otherwise stated.

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Yang, J., Wang, H., Geng, C. et al. Advances in intelligent diagnosis methods for pulmonary ground-glass opacity nodules. BioMed Eng OnLine 17, 20 (2018). https://doi.org/10.1186/s12938-018-0435-2

Download citation

  • Received:

  • Accepted:

  • Published:

  • DOI: https://doi.org/10.1186/s12938-018-0435-2

Keywords

BioMedical Engineering OnLine

ISSN: 1475-925X

Contact us