As precursors of lung adenocarcinoma, GGNs undergo progression, similar to AAHs, AISs, MIAs, and IACs. Without invasion of the lung interstitium and lymph nodes, limited resections, such as wedge resections and segmentectomies, are appropriate for PIAs, including AAHs and AISs. For MIAs and IACs, some cases have shown invasion of the local lymph nodes, making standard lobectomies necessary [
13‐
16]. Detecting potentially invasive malignant changes in GGNs can be challenging. To our knowledge, our use of an analysis of the M-P and V(M-P) parameters to determine GGN substructural features and changes to differentiate pre-invasive lesions from invasive ones is unprecedented. Traditionally, such differentiations have relied on VDTs or the average CT numbers for whole GGNs.
Measurement variability
In this study, Bland-Altman analyses were used to evaluate potentially malignant GGNs. Measurements of the maximum and peak CT numbers using automatic 3D-reconstruction post-processing software resulted in low inter- and intra-observer CVs. The low variability resulted in a significantly improved ability to detect interior growth using CT numbers in different areas of the nodule, relative to overall descriptions of the diameter or volume of a subgroup of malignant GGNs. Because of the slow growth typical of GGNs, volume or density changes can be subtle, emphasizing the need for a precise measurement method. For solid nodules, Revel et al. concluded that two-dimensional measurements are unreliable [
17]. As has been previously demonstrated for solid nodules, 3D volume measurements have lower intra- and inter-observer variabilities than do two-dimensional diameter measurements [
18].
As described in Fig.
3, the intra-observer CVs of the MAX and PEAK values were as small as 0.002 and 0.006, respectively; the CVs increased to 0.011 and 0.044, respectively, for M-P and V(M-P) values. The relatively higher CV for M-P may have resulted from a cumulative or aggregative effect of variations in both the MAX and PEAK values. Furthermore, the CV for V(M-P) accumulated the variations in MAX, PEAK, and M-P. Secondly, all traditional CT parameters were measured using the same tube voltages, tube currents, and section thicknesses. Therefore, the algorithm in which the density of the same type of tissue was measured on the CT screen did not change significantly. Our proposed parameters were post-processing data that were subjectively measured by radiologists. Most importantly, as the parameters created in this study were new and not proficiently or regularly measured by our radiologists during clinical work, further research investigating more precise measuring methods and strategies needs to be completed.
Difference between maximum and peak CT number (M-P)
We realized that GGNs are not homogeneous lesions, based on CT numbers, regardless of their subjectively judged subtype purity. In the presented study, the peak (PEAK) and maximum CT (MAX) numbers for nodules were combined to investigate the sub-structural densities of GGNs in our study. PEAK refers to the CT number associated with the greatest number of pixels in a nodule’s CT number histogram; these values are most likely exhibited in the main body of the GGN. Similarly, MAX refers to the CT numbers with the highest density and that are most likely located in the core, or solid part, of the GGN. Thus, the M-P values for mixed GGNs (mean, 736.43 ± 182.92 HU) were significantly larger (
p < 0.001) than those for pure GGNs (mean, 418.36 ± 229.22 HU). Therefore, M-P values reflect the degree of homogeneity between the core and periphery of a GGN. Additionally, as illustrated in Fig.
7, when the TV of a GGN increases, the density of the principal structure more closely approximates that of normal lung parenchyma. This results in an increasing density difference between the solid core and the main body of the nodule increasing as well.
In the present study, the M-P values for PIAs (mean, 408.67 ± 150.23 HU) were significantly smaller (p = 0.001) than those for IACs (mean, 667.43 ± 243.40 HU). As suggested above, PIAs would be expected to be more homogeneous than IACs. During the progression of GGNs and lung adenocarcinoma, AAHs and AISs start proliferating so that their size and their ability to be distinguished from normal cells decrease. Using a Spearman correlation analysis, a positive coefficient of 0.856 (p < 0.001) was obtained between M-P values and the TV of GGN nodules. Therefore, PIA lesions tend to be smaller than nodules and exhibit density differences between the solid centers and the peripheral main bodies. IACs have larger volumes and are more invasive; they also have increased solid center densities and higher densities in the peripheral areas of the nodules. Additionally, an AUC of 0.810 simultaneously suggests both a favorable specificity and a favorable sensitivity, when an M-P value approaches the cut-off point of 489.5 HU, for distinguishing PIAs from IACs. If the preoperative M-P value for a GGN is larger than 489.5 HU, there is a greater possibility of the lesion being IAC. Otherwise, it could undergo further follow-up to achieve a more accurate assessment.
Lee et al. found that the average CT number of a GGN could be used to discriminate between an invasive lesion and a non-invasive one at −472 HU [
19]. Similarly, Tamura et al. used the average CT number to evaluate GGN stability, suggesting that CT number cutoffs of −634.9 ± 15.3 HU and −712.1 ± 14.1 HU represented growing and stable GGNs, respectively [
20]. Some authors have already investigated the average CT number for GGNs that allow pathologic differentiation. Although others have studied the histogram peak CT numbers of GGNs, they have shown great interest in peak patterns, 5 to 95th percentile CT numbers, skewedness, and kurtosis [
21‐
30]; furthermore, all of these studies focused on the CT number for the whole GGN.
Average V(M-P)
Follow-up is an important approach for assessing the pathological quality of GGNs. For example, VDT has been studied as an effective indicator for diagnosing pulmonary nodules. Oda’s research suggested that VDTs of 859.2 ± 428.9, 421.2 ± 228.4, and 202.1 ± 84.3 day were diagnostic of AAH, bronchioloalveolar carcinoma (BAC), and IAC, respectively [
8]. Nevertheless, during a follow-up interval, the volume of most GGN lesions, especially for malignant ones, increases while the density also changes. If the VDT is used to evaluate the progression of the GGN exterior during a follow-up interval, then the use of V(M-P) should achieve a similar goal for the interior of the lesion. In the present study, the V(M-P) values of PIA lesions (6.92 ± 5.86 HU/day) were significantly larger (
p = 0.04) than those for IAC lesions (2.36 ± 6.86 HU/day). The absence of a negative sign before the results suggests increasing heterogeneity.
During the follow-up of PIA lesions, the proliferation of pre-invasive carcinoma cells is restricted to small lesions that do not extend outwards. AAH was defined as a lesion with a well-defined boundary, produced by the proliferation of mildly to moderately atypical type II pneumocytes or Clara cells lining the alveolar walls and respiratory bronchioles. Gaps are usually seen between the cells, which consist of rounded, cuboidal, low columnar or “peg” cells with round to oval nuclei. For AIS, localized small adenocarcinomas with growth restricted to neoplastic cells along preexisting alveolar structures (lepidic growth), there is a lack of stromal, vascular, or pleural invasion [
31]. Therefore, the interstitium is normal, without disordered structures or invasive malignant cells. Furthermore, the lung parenchyma, like the alveolar epithelium, continues to proliferate during the follow-up interval. Since only the small cells were growing rapidly, the density of the solid core area, on the CT screen, increased to higher values than did the main pixels associated with the nodule. Larger changes in the differences between the center and peripheral CT numbers were found in PIAs during the follow-up.
Malignant cells in IACs proliferated from multiple points throughout the nodule, invading the interstitium and filling it with mucous and isolated tumor cells so that, over time, the main part of the GGN showed an increasing density rather than only a solid core. Consequently, the V(M-P) values for IACs are remarkably smaller than those for PIAs, over the duration of the follow-up interval. An AUC of 0.805 was obtained, using the ROC assessment, for V(M-P) values allowing a differential diagnosis between PIAs and IACs, with a threshold of 11.01 HU/day. Accordingly, during the follow-up, if a GGN V(M-P) value larger than 11.01 HU/day is observed, PIA would be strongly suggested, rather than IAC.
Study Strengths and Limitations
All measurements used in our study were obtained automatically using 3D procedure software on a commercially available workstation. Quantitative 3D measurements of small pulmonary nodules, on CT images, are attracting increased attention. Such models can accurately measure parameters, such as the mean diameter and volume as well as all of the variables associated with CT numbers, and can present the information in an intuitive manner. Automatic segmentation software may minimize observer differences and also shorten the evaluation times; and new computer algorithms that are better suited for this task have also been developed. Most major CT vendors have included a GGN segmentation option in their pulmonary evaluation software.
Additionally, previous researches referring to GGN substructure was confined to simple classifications of subjectively assessed pure and mixed subtypes. The present study focused on the internal quantitative features of GGNs by comparing differences between the densities of the solid cores and the main bodies, as well as their changes during the follow-up. This was fortunate, as it proved effective for differentiating the types. Given GGN homogeneity, internal density differences provide information regarding oncological behavior, in addition to VDTs.
We compared the diagnostic capabilities of M-P and V(M-P) values with PEAK and V-PEAK reported by ourselves previously [
32]. We found that the AUCs in the ROC analysis of different parameters are very close to each other, notwithstanding a slightly higher value of PEAK. Nevertheless, after we concluded the several advantages and disadvantages of these two parameters, we believed that the M-P and V(M-P) characteristics, described in this present study, are superior to the PEAK and V-PEAK published at BJR. Our reasons are listed as follows: 1). According to other researches, such as Yasuhisa Ohde’s, the proportion of consolidation to GGO on high resolution CTs, at the respective maximum dimensions, was the best predictor of non-invasive peripheral lung adenocarcinoma [
33]. Therefore, M-P takes the solid part of GGN into consideration to evaluate pathological properties of nodules, given that M-P indicates the differences between the CT numbers for the solid parts and of peripheral parts in most cases. However, evaluation by only PEAK, involved in previous study published at BJR, just represents and descripts either peripheral ground-glass part (when proportion of consolidation less than 50%) or solid part (when proportion of consolidation more than 50%), omitting part radiological information of GGN during diagnosis and evaluation. 2). All of previous studies involving radiological diagnoses of GGN treated it as a whole entity. Unprecedentedly, in the present study, we created these two particular indexes, M-P and V(M-P), from a new perspective of substructure of GGNs to evaluate its radiological and pathological properties. 3). With regarding to mix GGN, there must be two peaks in CT histogram when we measure PEAK. However, particularly, when consolidation of it equals to 50%, these two peaks would be the same height but different PEAK value, which makes it difficult and inappropriate to use PEAK only. 4). Additionally, PEAK would dramatically change when solid part of GGN near to 50% or so, affecting comparison between different individuals and different values of itself during follow-up. 5). Speaking of comparison between individuals, the same PEAK could be either CT number of solid part in mix GGN or peripheral ground-glass part in pure GGN but neglect different radiological and pathological properties.
Our study had several limitations. First, the sample size was relatively small as some of the patients with GGN nodules were followed-up outside of our hospital or only came to our hospital for surgical therapy. Second, considering that benign lesions changed greatly or disappeared from CT scans after follow-up, we only collected information on malignant nodules during this study. Third, the V(M-P) values, obtained using an equation similar to that used for calculating arithmetic means, only captured the average growth velocity of M-P values over the follow-up period. However, the rate of change for GGN features varies constantly. Lastly, as for Fig.
7, we just only analyzed the PIA group corresponding to ROC analysis because the relativity between total volume of invasive adenocarcinoma and its density is not such significantly high, considering its density in CT screen would not change so sensitive after advancing into invasive adenocarcinoma.