Semi-automated segmentation of a non-small cell lung cancer nodule. The contours of the segmented nodule were superimposed on transverse, coronal, and sagittal CT images. (A) Three-dimensional view of the nodule. (B) Transverse CT image. (C) Sagittal CT image. (D) Coronal CT image. The size of the nodule was enlarged for display purposes. The segmentation process for this nodule did not require any manual operations other than the selection of the nodule position.
CT histogram of a subsolid nodule. The frequency axis presents a value normalized by the total number of voxels inside the segmented nodule.
Flow chart for the visual assessment of CT histograms for the five-category (α, β, γ, δ, and ɛ) classification. Y and N represent “yes” and “no,” respectively.
Transverse CT images for examples of the five classification types. (a) Type α. (b) Type β. (c) Type γ. (d) Type δ. (e) Type ɛ. Each transverse CT image shows the largest cut surface of the nodule.
CT histograms for the five categories shown in Fig. 5. (a) Type α. (b) Type β. (c) Type γ. (d) Type δ. (e) Type ɛ. In each CT value histogram for the segmented nodule, the frequency was normalized by the total number of voxels inside the segmented nodule.
Decision tree classification for the five-category (α, β, γ, δ, and ɛ) classification based on a classification and regression tree (CART) using a two-feature combination of the frequency of the highest peak of the CT histogram and the 90th percentile obtained from the CT histogram. The ellipsoids show the decision nodes for the features used in the node and the cutoff level. The boxes represent the terminal nodes of the five-category classification. T1, −44.5 HU; T2, −350.0 HU; T3, 0.0611; T4, −527.5 HU. Y and N represent “yes” and “no,” respectively.
Scatter plot of NSCLC nodules in the feature space that were used in the decision tree classification of the five-category classification. The colored space shows a recursive binary partitioning of the input feature space along with the corresponding tree structure presented in Fig. 7. The arrows denote misclassification cases obtained by the tenfold stratified cross-validation, compared with the gold standard.
A Bland-Altman plot showing the interobserver variability for the nodule volume.
A Bland-Altman plot showing the variability for the nodule volumes measured from two repeated segmentation tasks by Observer 1. SD, standard deviation.
Nodule-size distribution in nodule segmentation result 1. The nodule diameter was calculated from the volume of the segmented nodule assuming that the nodules were spherical. Black and white bars denote, respectively, agreement and disagreement between two repeated classification tasks by Observer 1.
Relapse-free survival curves of patients with non-small cell lung cancer, organized by CT histogram type.
Response frequency of the two observers who evaluated a pulmonary nodule as α, β, γ, δ, or ɛ type by visual assessment of the CT histogram obtained from nodule segmentation result 1.
Response frequency of the two observers who evaluated a pulmonary nodule as α, β, γ, δ, or ɛ type by visual assessment of the CT histogram obtained from nodule segmentation result 2.
Nodule classification and characteristics of non-small cell lung cancer.
Univariate and multivariate analyses of prognostic factors.
Article metrics loading...
Full text loading...