Skip to main page content
Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2016 Dec;11(12):2120-2128.
doi: 10.1016/j.jtho.2016.07.002. Epub 2016 Jul 13.

Predicting Malignant Nodules From Screening CT Scans

Free PMC article

Predicting Malignant Nodules From Screening CT Scans

Samuel Hawkins et al. J Thorac Oncol. .
Free PMC article

Erratum in

  • Erratum.
    J Thorac Oncol. 2018 Feb;13(2):280-281. doi: 10.1016/j.jtho.2017.09.1959. J Thorac Oncol. 2018. PMID: 29425613 No abstract available.


Objectives: The aim of this study was to determine whether quantitative analyses ("radiomics") of low-dose computed tomography lung cancer screening images at baseline can predict subsequent emergence of cancer.

Methods: Public data from the National Lung Screening Trial (ACRIN 6684) were assembled into two cohorts of 104 and 92 patients with screen-detected lung cancer and then matched with cohorts of 208 and 196 screening subjects with benign pulmonary nodules. Image features were extracted from each nodule and used to predict the subsequent emergence of cancer.

Results: The best models used 23 stable features in a random forests classifier and could predict nodules that would become cancerous 1 and 2 years hence with accuracies of 80% (area under the curve 0.83) and 79% (area under the curve 0.75), respectively. Radiomics outperformed the Lung Imaging Reporting and Data System and volume-only approaches. The performance of the McWilliams risk assessment model was commensurate.

Conclusions: The radiomics of lung cancer screening computed tomography scans at baseline can be used to assess risk for development of cancer.

Keywords: Computed tomography; Lung cancer; Machine learning; Prediction; Radiomics; Screening.


Figure 1
Figure 1. Flowchart of cohorts
Both Cohorts 1 and 2 had a nodule-positive/cancer negative screen at time 0. Cohort 1 had a nodule positive screen at Time 1, of which 104 were diagnosed with a screen-detected lung cancer, SDLC. These were demographically matched to subjects with benign pulmonary nodules, bPN, and the same screening history. 208 bPN-1 were identified, and of these, 176 were successfully segmented. Cohort 2 had a nodule-positive/cancer negative screen at time 1, followed by a nodule- positive screen at Time 2, of which 92 had SDLC. These were demographically matched to 184 bPN subjects, of which 152 were successfully segmented. Segmentation errors are presented in Supplemental Table 1.
Figure 2
Figure 2. Images from SDLC and bPN at T0 and T1
The top images are from a patient with a benign pulmonary nodule, bPN, in cohort 1. The bottom images are from a patient with a screen-detected lung cancer, SDLC group, in cohort 1. The T0 scans appear similar to the eye, and growth can clearly be seen on the T1 SDLC scan, relative to no growth of the T1 bPN scan. Select radiomic features from the T0 scans that discriminated the groups are shown in the text boxes.
Figure 3
Figure 3. Binary classifier prediction
Receiver operator characteristics (ROC) curves of risk scores for McWilliams, our Random Forests based approach, and volume are shown (see text for details). McWilliams had an area under the ROC of 0.67 and volume had an AUC of 0.74, whereas the radiomics classifier using Random Forests had an AUC of 0.87.

Comment in

Similar articles

See all similar articles

Cited by 43 articles

See all "Cited by" articles

Publication types