Accurate diagnosis of lung adenocarcinoma type could efficiently improve the appropriate surgery and medication strategy, and hence benefit the therapy outcomes. Here, we tend to investigate novel features to make accurate classification among AIS, MIA, IAC. We found that age is a contributor to IAC development and smoking is positively correlated with the occurrence of MIA. Importantly, the unique features in different groups identify by TSCT had diagnosis value for lung adenocarcinoma.
Our results are consistent with previous studies. Shi et al revealed that there existed huge differences in the CT measurements of solitary GGNs between AIS-MIA and IAC [7]. Different from our study, they combined the AIS and MIA into one group. Their work also pointed out that nodule types, cystic appearances, diameters and CT values were different between groups, despite no differences were found in shapes, margins, bronchus sign, pleural indentation. These inconsistencies maybe due to different grouping method and different sample size (their sample size was much smaller than ours). In contrast, Another research showed that the unique feature of shape and margin could be used to discriminate different types of carcinoma types([8] [9]). Besides, patients in IAC group showed higher frequency of air bronchogram and pleural indentation, and CT values (HU) than AIS-MIA group [2]. Consistently, homma et al recently found that malignant signs could include lobulation, spiculation, vascular convergence and pleural indentation in invasive mucinous adenocarcinoma [10]. Although their sample size was small, their conclusion highly supported our results.
The correlation between age, smoking and lung adenocarcinoma types is controversial. We here believe that age is a contributor to IAC development, and smoking is positively correlated with MIA formation. Consistently, one group also found that ages are comparable between AIS and MIA patients but showed correlation with the occurrence of IAC [11]. In contrast, it was also reported that age and smoking showed no obvious correlation with lung adenocarcinoma types[12]. Our results implied that more treatment should be adopted to inhibit tumor invasion for elder patients. As for the features of smoking, we paradoxically found that smoking ratio increased in MIA compared with AIS but dropped in IAC group. A pooled estimation only showed weak effect of secondhand tobacco smoke exposure on AIS/MIA incidence [13], and another study also observe no correlation between smoking history and invasive carcinoma [14]. However, it was also reported that tobacco exposure may accelerate malignant progression [15], and smoking is not a major cause of AIS but play a role in the progression from AIS to the invasive adenocarcinoma still with AIS features [16].
We found that patients in ICA showed unique features of higher solid or GGO components and diameter of GGN. The solid components can be firosis, invasive component, collapsed alveolar or other parenchyma, which was related to fibrotic focus, alveoli collapse and tumor cell proliferation [17]. It was believed that GGO-dominant tumors could be identified as minimally invasive and with recent 5-year survival rates exceeding 90% [18]. In contrast, the majority of AIS and MIA cases had a solid proportion less than 50%. As Fig. 2 showed, the solid components in MGGN is the most powerful variable to identify IAC patients. The CT value of solid or GGO component is also impacted by air cavity densities, which include air bronchograms and vacuole sign. Air bronchograms in MIA commonly present distortion and extension in the GGO component. In this study, we found that air bronchograms in GGO component was dramatically higher in the MIA group than the other two groups, and thus showed an obviously predictive value to MIA. In addition, vacuole sign is another factor belonging to air cavity densities, as well as a useful indicator for IAC in our study. There is no existing evidence about the diagnostic role the vacuole sign in adenocarcinoma recognition. Together, besides CT values of solid or GGO components and diameters, the vacuole sign and air bronchograms might be supplemental features in predicting the lung adenocarcinoma types.
More importantly, we analyzed the diagnostic of GGO components, diameters, the vacuole sign and air bronchograms by receiver operating characteristic (ROC) analyses to convincingly testified the correlation between those parameters and IAC types. Further, we also adopted a linear regression model, C-Maker, to comprehensively evaluated the role played by 13 variables. Noteworthily, we draw a ROC curve using C-marker, which exhibited an AUS as high as 0.982. Its cutoff value for IAC was 0.287, with a sensitivity of 0.935 and specificity of 0.977, indicating that those variables was applicable in the determination of lung carcinoma type.
This study has some limitations. First, the sample size of three groups was still not very large. Or else we can perform a principal component analysis and then choose the most conclusive parameters to carry out machine learning, which may predict the probability of each subtype for any specific patient. Second, we demonstrated smoking is positively correlated with MIA formation, but the mechanism underlying the higher smoking rate in MIA group while IAC paradoxically had a lower rate is still confusing. Third, smoking is an important variable related to adenocarcinoma clarifications (especially for MIA), but the smoking habits of the smokers were not detailed or presented in pack-years. If more details were recorded, more convincing correlations might be probed.