AI for Multistructure Incidental Findings and Mortality Prediction at Chest CT in Lung Cancer Screening

Abstract

Background Incidental extrapulmonary findings are commonly detected on chest CT scans and can be clinically important. Purpose To integrate artificial intelligence (AI)-based segmentation for multiple structures, coronary artery calcium (CAC), and epicardial adipose tissue with automated feature extraction methods and machine learning to detect extrapulmonary abnormalities and predict all-cause mortality (ACM) in a large multicenter cohort. Materials and Methods In this post hoc analysis, baseline chest CT scans in patients enrolled in the National Lung Screening Trial (NLST) from August 2002 to September 2007 were included from 33 participating sites. Per scan, 32 structures were segmented with a multistructure model. For each structure, 15 clinically interpretable radiomic features were quantified. Four general codes describing abnormalities reported by NLST radiologists were applied to identify extrapulmonary significant incidental findings on the CT scans. Death at 2-year and 10-year follow-up and the presence of extrapulmonary significant incidental findings were predicted with ensemble AI models, and individualized structure risk scores were evaluated. Area under the receiver operating characteristic curve (AUC) analysis was used to evaluate the performance of the models for prediction of ACM and extrapulmonary significant incidental findings. The Pearson χ² test and Kruskal-Wallis rank sum test were used for statistical analyses. Results A total of 24 401 participants (median age, 61 years [IQR, 57-65 years]; 14 468 male) were included. In 3880 of 24 401 participants (16%), 4283 extrapulmonary significant incidental findings were reported. During the 10-year follow-up, 3389 of 24 401 participants (14%) died. CAC had the highest feature importance for predicting the three study end points. The 10-year ACM model demonstrated the best AUC performance (0.72; per-year mortality of 2.6% above and 0.8% below the risk threshold), followed by 2-year ACM (0.71; per-year mortality of 1.13% above and 0.3% below the risk threshold) and prediction of extrapulmonary significant incidental findings (0.70; probability of occurrence of 25.4% above and 9.6% below the threshold). Conclusion A fully automated AI model indicated extrapulmonary structures at risk on chest CT scans and predicted ACM with explanations. ClinicalTrials.gov Identifier: NCT00047385 © RSNA, 2024 Supplemental material is available for this article. See also the editorial by Yanagawa and Hata in this issue.

Graphical abstract

Figure 1:

Schematic shows an overview of the study. The fully automated artificial intelligence model integrates multistructure segmentation, quantitative image analysis, and epicardial adipose tissue and coronary artery calcium to rank structures according to risk on low-dose chest CT scans. convLSTM = convolutional long short-term memory, HU = Hounsfield unit, nnUNet = no-new-Net.

Figure 2:

Flowchart shows study inclusion and dichotomization of model performance validation. DICOM = Digital Imaging and Communications in Medicine.

Figure 3:

Model performance and feature importance for all-cause mortality (ACM) end points in participants undergoing low-dose chest CT during the National Lung Screening Trial. Left: Receiver operating characteristic curves show performance for all imaging, demographic characteristics (age, sex, race and ethnicity, body mass index), and a combination of all imaging and demographics for 10-year (A) and 2-year (B) ACM, with red dots indicating significant incidental findings as a predictor of ACM. Right: Horizontal bar graphs show per-structure feature importance scoring for 10-year (A) and 2-year (B) ACM. For a comparison between model performance and standard aortic measurements, see Table S8. * = P < .001. AUC = area under the receiver operating characteristic curve, NS = not significant.

Figure 4:

Examples of images in participants who underwent low-dose chest CT for lung cancer screening with cardiovascular structures identified at high risk of mortality, with red indicating the structure with the highest risk score, pink indicating structures with an increased risk score, and blue indicating structures with a low risk score. (A) Axial (left) and sagittal (middle) low-dose CT scans, with corresponding deep learning structure segmentation (bottom), and three-dimensional reconstructions (right) of segmented and ranked structures in a 64-year-old female participant show an aortic aneurysm (cross-sectional diameter of ascending aorta is 51 mm and descending aorta is 43 mm). (B) Waterfall plot in the same participant shows the aorta (red arrow) contributed most to death probability compared with other segmented structures. The participant died of a thoracic aortic aneurysm rupture 5 years after trial randomization. (C) Axial low-dose CT images (left, middle), with corresponding deep learning structure segmentation (bottom), and three-dimensional reconstructions (right) of segmented and ranked structures in a 62-year-old female participant show coronary artery calcifications in the right coronary artery and left descending artery. (D) Waterfall plot in the same participant shows coronary artery calcium (red arrow) contributed most to predicting mortality compared with other segmented structures. The participant died 2 years 2 months after randomization because of acute myocardial infarction.

Figure 5:

Examples of images in participants who underwent low-dose chest CT for lung cancer screening with noncardiovascular structures identified at high risk of mortality, with red indicating the structure with the highest risk score, pink indicating structures with an increased risk score, and blue indicating structures with a low risk score. (A) Axial (left) and sagittal (middle) low-dose CT scans, with corresponding deep learning structure segmentation (bottom), and three-dimensional reconstructions (right) of segmented and ranked structures in a 65-year-old male participant show thickening of the esophageal wall. (B) Waterfall plot in the same participant shows the esophagus (red arrow) contributed most to predicting mortality compared with other segmented structures. The participant died 4 years after randomization due to an unspecified malignant neoplasm of the esophagus. (C) Axial (left) and coronal (middle) low-dose CT scans, with corresponding deep learning structure segmentation (bottom), and three-dimensional reconstructions (right) of segmented and ranked structures in a 55-year-old male participant show decreased liver density (mean, 32 HU). (D) Waterfall plot in the same participant shows the liver (red arrow) contributed most to predicting mortality compared with other segmented structures. The participant died 4 years 7 months after randomization due to unspecified liver disease.