Purpose To identify the features at CT that are predictive of spread through air spaces (STAS) in surgically resected lung adenocarcinomas. Materials and Methods For this retrospective study, presence of STAS was evaluated in 948 consecutive patients who underwent surgical resection for lung adenocarcinoma from April 2015 to December 2016. Patients who were positive for STAS and negative for STAS were matched at a ratio of 1:2 by using patient variables (age, sex, and smoking status). CT features (ie, percentage of solid component, maximum diameter of solid component, lesion density, location, margin, shape, pseudocavity, calcification, central low attenuation, ill-defined peripheral opacity, air bronchogram, satellite lesions, and pleural retraction) were analyzed by using multivariable logistic regression and receiver operating characteristic curves. Results The final study population consisted of 276 patients (mean age, 59 years; age range, 32-78 years) including 129 men (mean age, 60 years; age range, 36-78 years) and 147 women (mean age, 59 years; age range, 32-78 years). Ninety-two patients were positive for STAS and 184 patients were negative for STAS. STAS was more common in solid tumors (71 of 92; 77%) than in part-solid (21 of 92; 23%) or ground-glass lesions (0 of 92; 0%) (P < .001). STAS was also associated with central low attenuation, ill-defined opacity, air bronchogram, and percentage of solid component (all P < .001). Percentage of solid component was an independent predictor of STAS (odds ratio, 1.06; 95% confidence interval: 1.03, 1.08) and a cut-off value of 90% showed a discriminatory power with a sensitivity of 89.2% and a specificity of 60.3%. Conclusion Percentage of solid component was independently associated with spread through air spaces in lung adenocarcinomas. © RSNA, 2018 Online supplemental material is available for this article.

Hepatocellular carcinoma (HCC) is one of the most common malignant tumors and a leading cause of cancer-related death worldwide. We propose a fully automated deep learning model to detect HCC using hepatobiliary phase magnetic resonance images from 549 patients who underwent surgical resection. Our model used a fine-tuned convolutional neural network and achieved 87% sensitivity and 93% specificity for the detection of HCCs with an external validation data set (54 patients). We also confirmed whether the lesion detected by our deep learning model is a true lesion using a class activation map.

No abstract