Machine learning can predict malignancy risk in patients with multiple pulmonary nodules
Click Here to Manage Email Alerts
A newly developed machine learning-based model successfully predicted risk for malignancy in patients with multiple pulmonary nodules, researchers reported in Clinical Cancer Research.
“The increasing detection rate of multiple pulmonary nodules has led to an emerging problem for lung cancer diagnosis,” Young Tae Kim, MD, PhD, professor in the department of thoracic and cardiovascular surgery at Seoul National University Hospital and the Seoul National University College of Medicine in South Korea, said in a press release. “Because many nodules are found to be benign either after long-term follow-up or surgery, it is important to carefully evaluate these nodules prior to invasive procedures.”
Kim and colleagues developed the machine learning-based model (PKU-M) to predict the probability of cancer using radiographic nodule characteristics and sociodemographic variables. The researchers then trained and validated the model the tool in multiple cohorts to estimate the malignant probability of multiple pulmonary nodules to guide physician decision making.
The model was first trained to predict malignancy in 1,739 pulmonary nodules in 520 patients who were treated at Peking University People’s Hospital from 2007 to 2018. The researchers reported that the model showed excellent discrimination (area under the curve, 0.909) and calibration (Brier score, 0.122) in this training cohort. The top predictive features of the model included nodule size, count and distribution and patient age, according to the release.
The model was then validated in 583 nodules from 220 patients who underwent surgery at six hospitals in China and Korea from 2016 to 2016. In this cohort, the model’s performance was similar to that in the training cohort (AUC, 0.89). When the researchers compared the model’s performance with four logistic regression-based models developed for the prediction of lung cancer, the PKU-M model outperformed the other four (P < .001 for comparison), according to the results.
Next, the researchers conducted a prospective comparison between the model, three thoracic surgeons, a radiologist and the RX artificial intelligence tool, which was previously established for lung cancer diagnosis. Performance was validated in 200 nodules in 78 patients at four hospitals in China from January to March 2019. Results again showed higher performance of the PKU-M model (AUC, 0.87) compared with the thoracic surgeons (AUC ranged from 0.73-0.79), radiologist (AUC, 0.75) and the RX artificial intelligence tool (AUB, 0.76), according to the results. The researchers reported the PKU-M model also outperformed clinicians, with a 14.3% increase in sensitivity and 7.8% increase in specificity.
Predictive tools for patients with multiple pulmonary nodules are limited, as currently available tools can predict malignancy in patients with single nodules, according to the researchers.
“Our prediction model, which was exclusively established for patients with multiple nodules, can help not only mitigate unnecessary surgery but also facilitate the diagnosis and treatment of lung cancer,” Kim said.
The researchers noted several limitations of the study. For one, the study was limited to Asian patients, so the results may not be generalizable to Western populations or other populations.