Longitudinal variability of CT imaging features for predicting pulmonary nodule invasiveness: A multicenter study

Area of Science:

• Radiology
• Oncology
• Medical Imaging

Background:

• Lung cancer screening with low-dose computed tomography (LDCT) can lead to overdiagnosis.
• Accurate prediction of invasive lung cancer from nodules is crucial for effective screening.

Purpose of the Study:

• To develop and validate a model predicting invasive lung cancer using longitudinal radiological features from multiple LDCT scans.
• To address and mitigate overdiagnosis in lung cancer screening.

Main Methods:

• Retrospective analysis of 628 patients with pulmonary nodules undergoing three LDCT scans.
• Development of a multivariate logistic model (MCT-ILC) incorporating nodule features and variability.
• External validation in two independent cohorts (n=252 and n=269) using AUC, calibration plots, and Decision Curve Analysis (DCA).

Main Results:

• The MCT-ILC model, using eleven factors including standard deviation of diameter variability (SD_diameter), showed high predictive performance.
• AUCs in testing cohorts were 0.912 and 0.906, outperforming models with only baseline features.
• The model demonstrated excellent calibration (Brier scores 0.091 and 0.078) and maintained positive net benefits across all thresholds via DCA.

Conclusions:

• The developed MCT-ILC model accurately assesses pulmonary nodule invasiveness.
• This model has the potential to significantly reduce overdiagnosis in lung cancer screening.
• The model effectively classifies patients, identifying over 90% of invasive nodules as high-risk.