Radiologic Features of Small Pulmonary Nodules and Lung Cancer Risk in the National Lung Screening Trial: A Nested Case-Control Study

Abstract

Purpose Using data and images from the National Lung Screening Trial (NLST), we extracted radiological features from small pulmonary nodules (SPNs) that did not meet the original criteria to be considered a positive screen and identified features associated with lung cancer risk. Methods We extracted radiological features from SPNs of baseline low-dose computed tomography screens that did not meet NLST criteria to be considered a positive screen. SPNs were identified for 73 incidence case patients that were diagnosed with lung cancer at either the first or second follow-up screen and for 157 control subjects who had had three consecutive negative screens. Multivariable logistic regression was used to assess the association between radiological features and lung cancer risk. All statistical tests were two-sided. Results Nine features were significantly different between case patients and control subjects. Backward-elimination followed by bootstrap re-sampling identified a reduced model of highly informative radiological features with an under the ROC curve of 0.932 (95% C] 0.88–0.96), specificity of 92.38% (95% CI 52.22–84.91), and sensitivity of 76.55% (95% CI 87.50–95.35) that included total emphysema score (OR=1.71; 95% CI 1.39–2.01), attachment to vessel (OR=2.41; 95% CI 0.99–5.81), nodule location (OR=3.25; 95% CI 1.09–8.55), border definition (OR=7.56; 95% CI 1.89–30.8), and concavity (OR=2.58; 95% CI 0.89–5.64). Conclusion We identified a set of clinically relevant radiological features that that can be easily scored in the clinical setting and may be of use to determine lung cancer risk among participants presenting with SPNs.

Artificial intelligence in cancer imaging: Clinical challenges and applications

“…34 A separate study was conducted to identify semantic features from small pulmonary nodules (less than 6 mm) to predict lung cancer incidence in the lung cancer screening setting and the revealed final model yielded an area under the curve of the receiver operating characteristic of 0.930 based on total emphysema score, attachment to vessel, nodule location, border definition, and concavity. 54 Although there was an imbalance between malignant and benign nodules in the aforementioned analyses, these studies provide compelling evidence for the utility of semantic features in lung cancer screening. As with nodules detected in the lung cancer screening setting, the standard of care for incidental pulmonary nodules lacks accurate decision tools for predicting malignancy versus benign disease and indolent versus aggressive behavior.…”

“…In a recent study, a model incorporating 4 quantitatively scored semantic features (short‐axis diameter, contour, concavity, and texture) conferred an accuracy of 74.3% to distinguish malignant from benign nodules in the lung cancer screening setting . A separate study was conducted to identify semantic features from small pulmonary nodules (less than 6 mm) to predict lung cancer incidence in the lung cancer screening setting and the revealed final model yielded an area under the curve of the receiver operating characteristic of 0.930 based on total emphysema score, attachment to vessel, nodule location, border definition, and concavity . Although there was an imbalance between malignant and benign nodules in the aforementioned analyses, these studies provide compelling evidence for the utility of semantic features in lung cancer screening.…”

Machine Learning Principles for Radiology Investigators

“…In radiology, bootstrapping has been used in a trial of Small Pulmonary Nodules and Cancer Risk in the National Lung Screening Trial (27). In that study, multivariate models were generated with findings validated by bootstrapping in a subset of 125 patients with a screening-detected lung cancer and prior negative CT screening studies, along with 250 case-matched controls with small 4À6 mm lung nodules.…”

Hypoxia-related radiomics predict immunotherapy response: A multi-cohort study of NSCLC

“…The second dataset 22,23 included 47 adenocarcinoma patients who underwent surgical resection as first course therapy at the Maastricht Radiation Oncology Clinic and had presurgical CT scans within 2 months prior to surgery. The third dataset included 234 patients 24,25 diagnosed with screen-detected incident lung cancers in the National Lung Screening Trial. The fourth dataset included 103 adenocarcinoma patients from the radiogenomics dataset that was described above.…”