Diagnostic performance for pulmonary adenocarcinoma on CT: comparison of radiologists with and without three-dimensional convolutional neural network.

OBJECTIVES: To compare diagnostic performance for pulmonary invasive adenocarcinoma among radiologists with and without three-dimensional convolutional neural network (3D-CNN). METHODS: Enrolled were 285 patients with adenocarcinoma in situ (AIS, n = 75), minimally invasive adenocarcinoma (MIA, n = 58), and invasive adenocarcinoma (IVA, n = 152). A 3D-CNN model was constructed with seven convolution-pooling and two max-pooling layers and fully connected layers, in which batch normalization, residual connection, and global average pooling were used. Only the flipping process was performed for augmentation. The output layer comprised two nodes for two conditions (AIS/MIA and IVA) according to prognosis. Diagnostic performance of the 3D-CNN model in 285 patients was calculated using nested 10-fold cross-validation. In 90 of 285 patients, results from each radiologist (R1, R2, and R3; with 9, 14, and 26 years of experience, respectively) with and without the 3D-CNN model were statistically compared. RESULTS: Without the 3D-CNN model, accuracy, sensitivity, and specificity of the radiologists were as follows: R1, 70.0%, 52.1%, and 90.5%; R2, 72.2%, 75%, and 69%; and R3, 74.4%, 89.6%, and 57.1%, respectively. With the 3D-CNN model, accuracy, sensitivity, and specificity of the radiologists were as follows: R1, 72.2%, 77.1%, and 66.7%; R2, 74.4%, 85.4%, and 61.9%; and R3, 74.4%, 93.8%, and 52.4%, respectively. Diagnostic performance of each radiologist with and without the 3D-CNN model had no significant difference (p > 0.88), but the accuracy of R1 and R2 was significantly higher with than without the 3D-CNN model (p < 0.01). CONCLUSIONS: The 3D-CNN model can support a less-experienced radiologist to improve diagnostic accuracy for pulmonary invasive adenocarcinoma without deteriorating any diagnostic performances. KEY POINTS: • The 3D-CNN model is a non-invasive method for predicting pulmonary invasive adenocarcinoma in CT images with high sensitivity. • Diagnostic accuracy by a less-experienced radiologist was better with the 3D-CNN model than without the model.

Lung Adenocarcinoma at CT with 0.25-mm Section Thickness and a 2048 Matrix: High-Spatial-Resolution Imaging for Predicting Invasiveness.

Background High-spatial-resolution (HSR) CT provides detailed information and clear delineation of lung anatomy and disease states. HSR CT may have high diagnostic performance for predicting invasiveness of lung adenocarcinoma. Purpose To examine the diagnostic performance of HSR CT in predicting the invasiveness of lung adenocarcinoma. Materials and Methods In this retrospective study, 89 consecutive patients with adenocarcinoma in situ (AIS), minimally invasive adenocarcinoma (MIA), or invasive adenocarcinoma (IVA) were included who underwent surgery for lung cancer between January 2018 and December 2019. All patients underwent HSR CT with 0.25-mm section thickness and a 2048 matrix. Two independent observers evaluated the images for the presence or absence of the following HSR CT findings: lobulation, spiculation, pleural indentation, vessel convergence, homogeneity of ground-glass opacity, reticulation, irregularity and centrality of solid portion, and air bronchiologram (irregularity, disruption, or dilatation). The total diameter (≤1.6 cm or >1.6 cm) and the longest diameter of the solid portion (≤0.8 cm or >0.8 cm) were evaluated. Logistic regression models were used to identify findings associated with MIA plus IVA. Receiver operating characteristic analysis was performed to determine diagnostic performance. Results Eighty-nine patients (mean, 69 years ± 11 [standard deviation]; 49 men) were evaluated. The size of the nodules with invasion was a mean of 2.5 cm ± 1.2. Univariable analysis revealed lobulation, spiculation, pleural indentation, irregular and central solid portion, air bronchiologram with disruption and/or irregular dilatation, and total and solid portion diameters as associated with MIA plus IVA (all, P < .05). After adjustment for age, sex, and pack-years of smoking, disruption of air bronchogram and solid portion diameter greater than 0.8 cm remained as predictors of invasiveness (P = .001 and P = .02, respectively). The diagnostic performance of these two findings combined were as follows: sensitivity of 97% (59 of 61 patients; 95% confidence interval: 94%, 100%) and specificity of 86% (19 of 22 patients; 95% confidence interval: 65%, 97%), with an area under the curve of 0.94. Conclusion Using high-spatial-resolution CT, disruption of air bronchiologram and a solid portion greater than 0.8 cm were independently associated with a greater likelihood of invasiveness in lung adenocarcinoma. © RSNA, 2020 Online supplemental material is available for this article. See also the editorial by Lynch and Oh in this issue.

Computed tomography (CT) findings in 88 neurofibromatosis 1 (NF1) patients: Prevalence rates and correlations of thoracic findings.

PURPOSE: To evaluate the prevalence rates and the correlations of thoracic computed tomography (CT) findings of neurofibromatosis 1 (NF1) in 88 patients. MATERIALS AND METHODS: Chest CT images of 88 NF1 patients were independently reviewed by three observers, and the CT findings were evaluated. If abnormal findings were present, their number, size, and distribution were recorded. The prevalence rate of each CT finding was calculated, and the correlations between CT findings were analyzed. RESULTS: Of the 88 cases, 13 were positive for cysts, 16 for emphysema, 8 for nodules, 8 for GGNs (ground glass nodules), 13 for mediastinal masses, 20 for scoliosis, 44 for subcutaneous nodules, and 34 for skin nodules. Cysts showed upper and peripheral dominant distributions. Regarding 13 mediastinal masses, 2 were diagnosed as malignant peripheral nerve sheath tumors (MPNSTs), 1 was diagnosed as primary lung cancer, 2 were diagnosed as lateral meningocele, 3 were diagnosed as neurofibromas, and the remaining 7 were considered neurofibromas. There was a significant correlation between the prevalence of subcutaneous nodules and that of skin nodules. Significant positive correlations were also seen between size and number, size and rate of central distribution, and number and rate of central distribution of cysts. CONCLUSION: Various CT findings were found in NF-1 patients, and the prevalence rates of subcutaneous and skin nodules were higher than other findings. Though the prevalence rates of subcutaneous nodules and skin nodules were significantly correlated, the other CT findings in NF-1 occurred independently. The number, size, and distribution of the cysts showed significant positive correlations with each other.

Paradoxical signal pattern of mediastinal cysts on T2-weighted MR imaging: phantom and clinical study.

PURPOSE: To evaluate the intracystic MRI (magnetic resonance imaging) signal intensity of mediastinal cystic masses on T2-weighted images. MATERIALS AND METHODS: A phantom study was performed to evaluate the signal intensity of a mediastinal cystic mass phantom (rubber balloon containing water) adjacent to a cardiac phantom pulsing at the rate of 60/min. T2-weighted images (sequence, fast spin echo [FSE] and single shot fast spin echo [SSFSE]) were acquired for the mediastinal cystic mass phantom. Further, a clinical study was performed in 33 patients (16 men, 17 women; age range, 19-85 years; mean, 65 years) with thymic cysts or pericardial cysts. In all patients, T2-weighted images (FSE and SSFSE) were acquired. The signal intensity of cystic lesion was evaluated and was compared with that of muscle. A region of interest (ROI) was positioned on the standard MR console, and signal intensity of the cystic mass (cSI), that of the muscle (mSI), and the rate of absolute value of cSI-mSI to standard deviation (SD) of background noise (|cSI-mSI|/SD=CNR [contrast-to-noise ratio]) were measured. RESULTS: The phantom study demonstrated that the rate phantom-ROI/saline-ROI was higher in SSFSE (0.36) than in FSE (0.19). In clinical cases, the degree of the signal intensity was higher in SSFSE than in FSE. The CNR was significantly higher in SSFSE (mean ± standard deviation, 111.0 ± 47.6) than in FSE (72.8 ± 36.6) (p<0.001, Wilcoxon signed-rank test). CONCLUSIONS: Anterior mediastinal cysts often show lower signal intensity than the original signal intensity of water on T2-weighted images. SSFSE sequence reduces this paradoxical signal pattern on T2-weighted images, which may otherwise cause misinterpretation when assessing cystic lesions.

Slow-growing primary marginal zone B-cell lymphoma arising in the chest wall in a patient without a history of tuberculosis.

A 57-year-old man with a 15-year history of a right chest wall mass lesion without follow-up for 5 years was admitted to our hospital without any symptoms or evidence of malignancy. On MRI, an additional small subcutaneous mass lesion was found. Histology of both lesions revealed marginal zone B-cell lymphomas. Primary chest wall marginal zone B-cell lymphomas are rare; this report discusses the case and the literature on chest wall lymphomas.