[Performance of Deep-learning-based Artificial Intelligence on Detection of Pulmonary Nodules in Chest CT].

BACKGROUND: The detection of pulmonary nodules is a key step to achieving the early diagnosis and therapy of lung cancer. Deep learning based Artificial intelligence (AI) presents as the state of the art in the area of nodule detection, however, a validation with clinical data is necessary for further application. Therefore, the aim of this study is to evaluate the performance of AI in the detection of malignant and non-calcified nodules in chest CT. METHODS: Two hundred chest computed tomography (CT) data were randomly selected from a self-built nodule database from Tianjin Medical University General Hospital. Both the pathology confirmed lung cancers and the nodules in the process of follow-up were included. All CTs were processed by AI and the results were compared with that of radiologists retrieved from the original medical reports. The ground truths were further determined by two experienced radiologists. The size and characteristics of the nodules were evaluated as well. The sensitivity and false positive rate were used to evaluate the effectiveness of AI and radiologists in detecting nodules. The McNemar test was used to determine whether there was a significant difference. RESULTS: A total of 889 non-calcified nodules were determined by experts on chest CT, including 133 lung cancers. Of them, 442 nodules were less than 5 mm. The cancer detection rates of AI and radiologists are 100%. The sensitivity of AI on nodule detection was significantly higher than that of radiologists (99.1% vs 43%, P<0.001). The false-positive rate of AI was 4.9 per CT and decreased to 1.5 when nodules less than 5 mm were excluded. CONCLUSIONS: AI achieves the detection of all malignancies and improve the sensitivity of pulmonary nodules detection beyond radiologists, with a low false positive rate after excluding small nodules.

[Growth Evaluation of Pulmonary Nodules on Chest CT].

Computed tomography (CT) follow-up of indeterminate pulmonary nodules and the quantification of growth characteristics are the commonly adopted strategy in clinical setting. The volume/mass doubling time can be used to quantify the growth velocity based on exponential growth model. Therefore, we reviewed the followed aspects on growth evaluation of pulmonary nodules on chest CT, including the growth model of lung cancer, the methods used for nodule growth quantification and the growth characteristics of different types of pulmonary nodules.

[Effect of Scanning and Reconstruction Parameters on Three Dimensional Volume and CT Value Measurement of Pulmonary Nodules: A Phantom Study].

BACKGROUND: The computed tomography (CT) follow-up of indeterminate pulmonary nodules aiming to evaluate the change of the volume and CT value is the common strategy in clinic. The CT dose needs to considered on serious CT scans in addition to the measurement accuracy. The purpose of this study is to quantify the precision of pulmonary nodule volumetric measurement and CT value measurement with various tube currents and reconstruction algorithms in a phantom study with dual-energy CT. METHODS: A chest phantom containing 9 artificial spherical solid nodules with known diameter (D=2.5 mm, 5 mm, 10 mm) and density (-100 HU, 60 HU and 100 HU) was scanned using a 64-row detector CT canner at 120 Kilovolt & various currents (10 mA, 20 mA, 50 mA, 80 mA,100 mA, 150 mA and 350 mA). Raw data were reconstructed with filtered back projection and three levels of adaptive statistical iterative reconstruction algorithm (FBP, ASIR; 30%, 50% and 80%). Automatic volumetric measurements were performed using commercially available software. The relative volume error (RVE) and the absolute attenuation error (AAE) between the software measures and the reference-standard were calculated. Analyses of the variance were performed to evaluate the effect of reconstruction methods, different scan parameters, nodule size and attenuation on the RPE. RESULTS: The software substantially overestimated the very small (D=2.5 mm) nodule's volume [mean RVE: (100.8%±28%)] and underestimated it attenuation [mean AAE: (-756±80) HU]. The mean RVEs of nodule with diameter as 5 mm and 10 mm were small [(-0.9%±1.1%) vs (0.9%±1.4%)], however, the mean AAEs [(-243±26) HU vs (-129±7) HU)] were large. The ANOVA analysis for repeated measurements showed that different tube current and reconstruction algorithm had no significant effect on the volumetric measurements for nodules with diameter of 5 mm and 10 mm (F=5.60, P=0.10 vs F=11.13, P=0.08), but significant effects on the measurement of CT value (F=34.79, P<0.001 vs F=156.14, P<0.001). CONCLUSIONS: An infinitesimally small errors of volumetric measurement of 5 mm or 10 mm nodule could achieved with very low current and ASIR reconstruction, suggesting a possibility of remarkable radiation dose reductions, while it is not applicable for 5 mm nodule. The attenuation acquired through three dimensional software has large measurement error and can not applied in clinical currently.
.

[Three Dimensional Volumetric Analysis of Solid Pulmonary Nodules on Chest CT: 
Cancer Risk Assessment].

BACKGROUND: The management of pulmonary nodules relies on cancer risk assessment, in which the only widely accepted criterion is diameter. The development of volumetric computed tomography (CT) and three-dimensional (3D) software enhances the clarity in displaying the nodules' characteristics. This study evaluated the values of the nodules' volume and 3D morphological characteristics (edge, shape and location) in cancer risk assessment. METHODS: The CT data of 200 pulmonary nodules were retrospectively evaluated using 3D volumetric software. The malignancy or benignity of all the nodules was confirmed by pathology, histology or follow up (>2 years). Logistic regression analysis was performed to calculate the odds ratios (ORs) of the 3D margin (smooth, lobulated or spiculated/irregular), shape (spherical or non-spherical), location (purely intraparenchymal, juxtavascular or pleural-attached), and nodule volume in cancer risk assessment for total and sub-centimeter nodules. The receiver operating characteristic (ROC) curve was employed to determine the optimal threshold for the nodule volume. RESULTS: Out of 200 pulmonary nodules, 78 were malignant, whereas 122 were benign. The Logistic regression analysis showed that the volume (OR=3.3; P<0.001) and the 3D margin (OR=13.4, 9.8; both P=0.001) were independent predictive factors of malignancy, whereas the location and 3D shape exhibited no total predictive value (P>0.05). ROC analysis showed that the optimal threshold for malignancy was 666 mm³. For sub-centimeter nodules, the 3D margin was the only valuable predictive factor of malignancy (OR=60.5, 75.0; P=0.003, 0.007). CONCLUSIONS: The volume and 3D margin are important factors considered to assess the cancer risk of pulmonary nodules. Volumes larger than 666 mm³ can be determined as high risk for pulmonary nodules; by contrast, nodules with lobulated, spiculated, or irregular margin present a high malignancy probability.

[The intra-observer variability of volumetric measurement of pulmonary nodules: comparison of two-dimensional and three-dimensional method].

BACKGROUND AND OBJECTIVE: Software oriented three-dimensional (3D) volumetric measurement of pulmonary nodules has been feasible in the follow-up of indeterminate pulmonary nodules, however, its value need a further validation. The purpose of this study is to retrospectively analyze the chest CT data of patients with pulmonary nodules to compare the intra-observer variability of 3D and two-dimensional (2D) volumetric measurement. METHODS: Eighty-six pulmonary nodules in chest CT scans of 79 subjects were retrospectively analyzed. One radiologist measured the nodules twice with a 7 days interval using 2D and 3D methods respectively. The maximal diameter (X), the perpendicular diameter (Y) on maximal cross sectional area of the nodule and the caudo-cranial diameter (Z) were measured and the volume was calculated by two models: spherical and elliptical model. The 3D measurements were acquired with semi-automated software with manual adjustment on unsatisfied nodule segmentation. Logistic regression analysis was performed to evaluate the effect of nodule location and morphology on 3D nodule segmentation. ANOVA and correlation test were used to evaluate the difference among three methods. Bland-Altman method was applied to quantify the intra-observer variability. RESULTS: Software achieved satisfied segmentation for 86.4% nodules. The irregular and juxtavacular nodules have significantly high odds rations (OR) of unsatisfied segmentation as 4.0, 4.5, respectively. The volume measured by three method was significantly different (F=6.5, P=0.012), while the repeated measurements did not led to significant difference (F=1.813, P=0.182). The Spearman correlation efficient between 3D volume and 2D volume with sphere and ellipsoid model was 0.97, 0.88. The 95% limits of agreement of RD between two repeated measurements were -14%-11.6%, -37.7%-39.9% and -39.8%-45.8% for 3D, 2D with elliptical model and spherical model, respectively. CONCLUSIONS: The 3D volume measurement of pulmonary nodules is more repeatable than 2D volume measurement. Unsatisfied segmentation can occurred on a small number of nodules, especially for irregular and juxtavascular nodules. For these nodules, the measurement of 3D diameters is recommended.

[Correlation of apparent diffusion coefficient with histologic type and grade of lung cancer].

BACKGROUND AND OBJECTIVE: The histologic type and grade of lung cancer are significant in assessing the biological behaviour, prognosis and therapeutic regimen. This study investigates the correlation of apparent diffusion coefficient (ADC) with histologic type and grade of lung cancer. METHODS: A total of 115 patients pathologically diagnosed with lung cancer were enrolled in this study. Magnetic resonance (MR)-DWI with a diffusion factor of 500 s/mm2 was performed and the ADC values of lesions were measured. The ADC values among the different histologic types and grades were compared with the values obtained using t-test and one-way ANOVA. The correlation between ADC values and the histologic grades was further evaluated with Spearman correlation coefficient. RESULTS: With a b value of 500 s/mm2, the ADC values significantly differed between small cell lung cancer (SCLC) and non-small cell lung cancer (NSCLC) (P = 0.017). The ADC values of SCLC were significantly different from those of squamous cell carcinoma and adenocarcinoma (P = 0.011, 0.001, respectively). A significant difference was observed in the ADC values among the different histologic grades of lung cancer (P = 0.003), and the ADC value was correlated with the histologic grades (rs = -0.272, P = 0.003). CONCLUSIONS: ADC value is significant for judgement of histologic type and grade of lung cancer before surgical operation: SCLC has a low ADC value, and ADC value is low for the tumor with poorly-differentiated grade.