A study published in Thorax found that the use of artificial intelligence (AI) may help reduce false positive rates in lung cancer screening.
Low-dose computed tomography (CT) is currently used in lung cancer screening of high-risk populations for early lung cancer diagnosis but 96% of detected nodules are false positives. So, researchers at the University of Pittsburgh developed a new lung cancer predictor after studying 218 individuals with lung cancer or benign nodules.
They used probabilistic graphical models to integrate demographics, clinical data, and low-dose CT features from 92 subjects (training cohort) from the Pittsburgh Lung Screening Study, a community-based research cohort of 3,642 current or former smokers recruited between 2002 and 2006.
All participants received a baseline low-dose CT scan, and most also received a follow-up scan the next year. Participants also completed a questionnaire about smoking history, underwent spirometry for pulmonary function testing, and provided a blood sample.
The validation cohort included 126 participants (44 cases and 82 controls).
New approach (probabilistic graphical models, which I cannot claim to understand…) used to develop nodule cancer prediction tool. Performed better than Brock/PanCan calculator parsimonious, but not full model. @ThoraxBMJ https://t.co/pPHW9J5MOe #lcsm #lungcancerscreening
— Lung Cancer Screening Feed (@lung_ca_screen) March 13, 2019
Positive outcome with AI
Nodule and vessel numbers and years since the subject quit smoking were sufficient to discriminate malignant from benign nodules. Comparison with existing predictors in the training and validation cohorts showed that incorporating low-dose CT scan features greatly enhances predictive accuracy and the machine-learning model improved cancer detection over existing methods, including the Brock parsimonious model (P<0.001). The researchers noted that the number of surrounding vessels, a feature not previously used in predictive models, significantly improves predictive efficiency.
LCCM could identify 28.3% of the benign nodules without risk of misclassifying cancer nodules, according to the authors.
The researchers said the findings should be validated in larger prospective studies.