Lung tumor segmentation using dual-coupling net with shape prior based on lung and mediastinal window images from chest CT images.

BACKGROUND: Volumetric lung tumor segmentation is difficult due to the diversity of the sizes, locations and shapes of lung tumors, as well as the similarity in the intensity with surrounding tissue structures. OBJECTIVE: We propose a dual-coupling net for accurate lung tumor segmentation in chest CT images regardless of sizes, locations and shapes of lung tumors.METHODSTo extract shape information from lung tumors and use it as shape prior, three-planar images including axial, coronal, and sagittal planes are trained on 2D-Nets. Two types of window images, lung and mediastinal window images, are trained on 2D-Nets to distinguish lung tumors from the thoracic region and to better separate the boundaries of lung tumors from adjacent tissue structures. To prevent false-positive outliers to adjacent structures and to consider the spatial information of lung tumors, pairs of tumor volume-of-interest (VOI) and tumor shape prior are trained on 3D-Net.RESULTSIn the first experiment, the dual-coupling net had the highest Dice Similarity Coefficient (DSC) of 75.7%, considering the shape prior as well as mediastinal window images to prevent the leakage of adjacent structures while maintaining the shape of the lung tumor, with 18.23% p, 3.7% p, 1.1% p, and 1.77% p higher DSCs than in the 2D-Net, 2.5D-Net, 3D-Net, and single-coupling net results, respectively. In the second experiment with annotations for two clinicians, the dual-coupling net showed outcomes of 67.73% and 65.07% regarding the DSC for each annotation. In the third experiment, the dual-coupling net showed 70.97% for the DSC.CONCLUSIONSThe dual-coupling net enables accurate segmentation by distinguishing lung tumors from surrounding tissue structures and thus yields the highest DSC value.

Prediction of Two-Year Recurrence-Free Survival in Operable NSCLC Patients Using Radiomic Features from Intra- and Size-Variant Peri-Tumoral Regions on Chest CT Images.

To predict the two-year recurrence-free survival of patients with non-small cell lung cancer (NSCLC), we propose a prediction model using radiomic features of the inner and outer regions of the tumor. The intratumoral region and the peritumoral regions from the boundary to 3 cm were used to extract the radiomic features based on the intensity, texture, and shape features. Feature selection was performed to identify significant radiomic features to predict two-year recurrence-free survival, and patient classification was performed into recurrence and non-recurrence groups using SVM and random forest classifiers. The probability of two-year recurrence-free survival was estimated with the Kaplan-Meier curve. In the experiment, CT images of 217 non-small-cell lung cancer patients at stages I-IIIA who underwent surgical resection at the Veterans Health Service Medical Center (VHSMC) were used. Regarding the classification performance on whole tumors, the combined radiomic features for intratumoral and peritumoral regions of 6 mm and 9 mm showed improved performance (AUC 0.66, 0.66) compared to T stage and N stage (AUC 0.60), intratumoral (AUC 0.64) and peritumoral 6 mm and 9 mm classifiers (AUC 0.59, 0.62). In the assessment of the classification performance according to the tumor size, combined regions of 21 mm and 3 mm were significant when predicting outcomes compared to other regions of tumors under 3 cm (AUC 0.70) and 3 cm~5 cm (AUC 0.75), respectively. For tumors larger than 5 cm, the combined 3 mm region was significant in predictions compared to the other features (AUC 0.71). Through this experiment, it was confirmed that peritumoral and combined regions showed higher performance than the intratumoral region for tumors less than 5 cm in size and that intratumoral and combined regions showed more stable performance than the peritumoral region in tumors larger than 5 cm.

CT quantification of the heterogeneity of fibrosis boundaries in idiopathic pulmonary fibrosis.

OBJECTIVES: To quantify the heterogeneity of fibrosis boundaries in idiopathic pulmonary fibrosis (IPF) using the Gaussian curvature analysis for evaluating disease severity and predicting survival. METHODS: We retrospectively included 104 IPF patients and 52 controls who underwent baseline chest CT scans. Normal lungs below - 500 HU were segmented, and the boundary was three-dimensionally reconstructed using in-house software. Gaussian curvature analysis provided histogram features on the heterogeneity of the fibrosis boundary. We analyzed the correlations between histogram features and the gender-age-physiology (GAP) and CT fibrosis scores. We built a regression model to predict diffusing capacity of carbon monoxide (DLCO) using the histogram features and calculated the modified GAP (mGAP) score by replacing DLCO with the predicted DLCO. The performances of the GAP, CT-GAP, and mGAP scores were compared using 100 repeated random-split sets. RESULTS: Patients with moderate-to-severe IPF had more numerous Gaussian curvatures at the fibrosis boundary, lower uniformity, and lower 10th to 30th percentiles of Gaussian curvature than controls or patients with mild IPF (all p < 0.0033). The 20th percentile was most significantly correlated with the GAP score (r = - 0.357; p < 0.001) and the CT fibrosis score (r = - 0.343; p = 0.001). More numerous Gaussian curvatures, higher entropy, lower uniformity, and 10th to 30th percentiles (p < 0.001-0.041) were associated with mortality. The mGAP score was comparable to the GAP and CT-GAP scores for survival prediction (mean C-indices, 0.76 vs. 0.79 vs. 0.77, respectively). CONCLUSIONS: Gaussian curvatures of fibrosis boundaries became more heterogeneous as the disease progressed, and heterogeneity was negatively associated with survival in IPF. KEY POINTS: • Gaussian curvature of the fibrotic lung boundary was more heterogeneous in patients with moderate-to-severe IPF than those with mild IPF or normal controls. • The 20th percentile of the Gaussian curvature of the fibrosis boundary was linearly correlated with the GAP score and the CT fibrosis score. • A modified GAP score that replaced the diffusing capacity of carbon monoxide with a composite measure using histogram features of the Gaussian curvature of the fibrosis boundary showed a comparable ability to predict survival to both the GAP and the CT-GAP score.

Author Correction: Inspiratory Lung Expansion in Patients with Interstitial Lung Disease: CT Histogram Analyses.

An amendment to this paper has been published and can be accessed via a link at the top of the paper.

Effect of CT Reconstruction Algorithm on the Diagnostic Performance of Radiomics Models: A Task-Based Approach for Pulmonary Subsolid Nodules.

OBJECTIVE: We investigated whether the diagnostic performance of machine learning-based radiomics models for the discrimination of invasive pulmonary adenocarcinomas (IPAs) among subsolid nodules (SSNs) was affected by the proportion of images reconstructed with filtered back projection (FBP) and model-based iterative reconstruction (MBIR) in datasets used for feature extraction. MATERIALS AND METHODS: This retrospective study included 60 patients (23 men and 37 women; mean age, 61.4 years) with 69 SSNs (54 part-solid and 15 pure ground-glass nodules). Preoperative CT scans were reconstructed with both FBP and MBIR. A total of 860 radiomics features were obtained from the entire nodule volume, and 70 resampled nodule datasets with an increasing proportion of nodules with MBIR-derived features (from 0/69 to 69/69) were prepared. After feature selection using neighborhood component analysis, support vector machines (SVMs) and an ensemble model were used as classifiers for the differentiation of IPAs. The diagnostic performances of all blending proportions of reconstruction algorithms were calculated and analyzed. RESULTS: The ROC AUC and the diagnostic accuracy of the radiomics models decreased significantly as the number of nodules with MBIR-derived features increased, and this relationship followed cubic functions (R2 = 0.993 and 0.926 for SVM; R2 = 0.993 and 0.975 for the ensemble model; p < 0.001). The magnitude of variation in AUC due to the reconstruction algorithm heterogeneity was 0.39 for SVM and 0.39 for the ensemble model. CONCLUSION: Inclusion of CT scans reconstructed with MBIR for radiomics modeling can significantly decrease diagnostic performance for the identification of IPAs.

Inspiratory Lung Expansion in Patients with Interstitial Lung Disease: CT Histogram Analyses.

This study aimed to evaluate inspiratory lung expansion in patients with interstitial lung disease (ILD) using histogram analyses based on advanced image registration between inspiratory and expiratory CT scans. We included 16 female ILD patients and eight age- and sex-matched normal controls who underwent full-inspiratory and expiratory CT scans. The CT scans were sequentially aligned based on the surface, landmarks, and attenuation of the lung parenchyma. Histogram analyses were performed on the degree of lung expansion (DLE) of each pixel between the aligned images in x-, y-, z-axes, and 3-dimensionally (3D). The overall mean registration error was 1.9 mm between the CT scans. The DLE3D in ILD patients was smaller than in the controls (mean, 17.6 mm vs. 26.9 mm; p = 0.023), and less heterogeneous in terms of standard deviation, entropy, and uniformity (p < 0.05). These results were mainly due to similar results in the DLEZ of the lower lungs. A forced vital capacity tended to be weakly correlated with mean (r2 = 0.210; p = 0.074), and histogram parameters (r2 = 0.194-0.251; p = 0.048-0.100) of the DLE3D in the lower lung in ILD patients. Our findings indicate that reduced and less heterogeneous inspiratory lung expansion in ILD patients can be identified by using advanced accurate image registration.

Improving the prediction of lung adenocarcinoma invasive component on CT: Value of a vessel removal algorithm during software segmentation of subsolid nodules.

PURPOSE: To evaluate the value of a vessel removal algorithm in segmentation of subsolid nodules by comparing the software solid component measurement on CT, before and after vessel removal, with the measurement of the invasive component on pathology in lung adenocarcinomas manifesting as subsolid nodules. MATERIALS AND METHODS: Between January 2014 and June 2015, 73 subsolid nodules with an invasive component of ≤10 mm on pathology were selected for analyses. For each nodule, semi-automated segmentation was performed by 2 radiologists and 3-dimensional (D) longest, axial longest and effective diameters of solid component were obtained from software, before and after using a vessel removal tool. These measurements were compared with the invasive component diameter on pathology using the paired t-test and Pearson's correlation test. RESULTS: Sixty-eight successfully segmented subsolid nodules were included. The mean maximal diameter of the invasive component on pathology was 4.6 mm (range, 0-10 mm). The correlation between software and pathology measurements was significant (p < 0.01) and the correlation after vessel removal (r = 0.49-0.54) was better than before vessel removal (r = 0.27-0.41). The mean measurement difference between solid component on CT and invasive tumor on pathology was significantly larger before vessel removal than after vessel removal in all measurements. The smallest mean measurement difference was obtained with 3D longest diameter of solid component after vessel removal in both readers (-0.26 mm to 0.10 mm), with no significant difference from pathology (p = 0.53-0.83). CONCLUSION: By adding a vessel removal algorithm in software segmentation of subsolid nodules, the prediction of invasive component in lung adenocarcinomas can be improved.

Ground-glass nodule segmentation in chest CT images using asymmetric multi-phase deformable model and pulmonary vessel removal.

We propose a ground-glass nodule (GGN) segmentation method that can separate solid component and ground-glass opacity (GGO) using an asymmetric multi-phase deformable model in chest CT images. First, initial solid component and GGO were extracted using intensity-based segmentation with histogram modeling. Second, the initial extracted regions were refined using an asymmetric multi-phase deformable model with modified energy functional and intensity-constrained averaging function. Finally, vessel-like structures are removed based on multi-scale shape analysis. In experiments, the segmentation accuracy of the entire GGN was evaluated using datasets from SNUH and LIDC/IDRI. The average DSC values of Seoul National University Hospital (SNUH) and Lung Image Database Consortium and Image Database Resource Initiative (LIDC/IDRI) were 0.85 ± 0.05 and 0.78 ± 0.07, respectively. The Pearson's correlation coefficient (r) between segmented volumes by the proposed method and manual segmentation was evaluated using SNUH dataset. The r of solid component, GGO, and entire GGN were 0.931, 0.875 and 0.907. Our experimental results show that the proposed method improves segmentation accuracy by applying the proposed asymmetric multiphase deformable model and pulmonary vessel removal.

Computer-aided classification of visual ventilation patterns in patients with chronic obstructive pulmonary disease at two-phase xenon-enhanced CT.

OBJECTIVE: To evaluate the technical feasibility, performance, and interobserver agreement of a computer-aided classification (CAC) system for regional ventilation at two-phase xenon-enhanced CT in patients with chronic obstructive pulmonary disease (COPD). MATERIALS AND METHODS: Thirty-eight patients with COPD underwent two-phase xenon ventilation CT with resulting wash-in (WI) and wash-out (WO) xenon images. The regional ventilation in structural abnormalities was visually categorized into four patterns by consensus of two experienced radiologists who compared the xenon attenuation of structural abnormalities with that of adjacent normal parenchyma in the WI and WO images, and it served as the reference. Two series of image datasets of structural abnormalities were randomly extracted for optimization and validation. The proportion of agreement on a per-lesion basis and receiver operating characteristics on a per-pixel basis between CAC and reference were analyzed for optimization. Thereafter, six readers independently categorized the regional ventilation in structural abnormalities in the validation set without and with a CAC map. Interobserver agreement was also compared between assessments without and with CAC maps using multirater κ statistics. RESULTS: Computer-aided classification maps were successfully generated in 31 patients (81.5%). The proportion of agreement and the average area under the curve of optimized CAC maps were 94% (75/80) and 0.994, respectively. Multirater κ value was improved from moderate (κ = 0.59; 95% confidence interval [CI], 0.56-0.62) at the initial assessment to excellent (κ = 0.82; 95% CI, 0.79-0.85) with the CAC map. CONCLUSION: Our proposed CAC system demonstrated the potential for regional ventilation pattern analysis and enhanced interobserver agreement on visual classification of regional ventilation.