Figure 1.
Transverse thin-section CT image showing manual segmentation of a part-solid nodule (PSN).
Segmentation of the PSN was manually conducted using an in-house software program and texture features of the nodules were automatically extracted and calculated by the program. One radiologist segmented the outer boundary of the whole PSN (A) and inner solid portion boundary (B).
Figure 2.
An example of texture analysis of a persistent PSN.
(A) Thin-section CT scan shows an 18 mm PSN (arrow) with fissural retraction in the right lower lobe in a 62-year-old male. (B) Texture analysis of the PSN shows high mean attenuation and low negative skewness (−305.5 Hounsfield units and −0.378, respectively). As this PSN was persistent, he underwent lobectomy and was diagnosed as having adenocarcinoma.
Figure 3.
An example of texture analysis of a transient PSN.
(A) Thin-section CT scan shows a 17 mm PSN (arrow) with an ill-defined margin in the left upper lobe in a 49-year-old female. (B) Texture analysis of this PSN shows a low mean attenuation and high positive skewness (−570.2 Hounsfield units and 0.856, respectively). The PSN disappeared at follow-up CT after one month.
Table 1.
Clinical features in 77 Individuals with Transient and Persistent PSNs.
Table 2.
CT features in 77 Transient and Persistent PSNs.
Table 3.
Texture Analysis Features in 77 PSNs.
Table 4.
The percentile CT numbers in 77 PSNs.
Table 5.
Results of Logistic Regression Analysis for Clinical, Thin-Section CT and Texture analysis of Transient and Persistent PSNs.
Figure 4.
C-statistic analysis of multiple logistic regression models in discriminating transient PSNs from persistent PSNs.
There were three combinations of independent predictors in the differentiation between transient and persistent PSNs. The highest area under the curve (AUC) was achieved for the combination of clinical, thin-section CT and texture analysis (AUC = 0.929 ? 0.0272). The AUC of clinical and thin-section CT predictors alone (AUC = 0.790 ? 0.0522) was not significantly different from the AUC of computer-aided quantified pixel value predictors alone (AUC = 0.831 ? 0.0503) (P = 0.598). However, the AUC of the combination of clinical, thin-section CT and computer-aided quantified pixel value predictors was significantly higher than that of either the clinical and thin-section CT or the texture analysis alone (P=0.004 and P=0.04). AUCs are shown as means ? standard deviations.