Multi-Window CT Based Radiological Traits for Improving Early Detection in Lung Cancer Screening

Area of Science:

• Radiology
• Oncology
• Medical Imaging Analysis

Background:

• Accurate prediction of lung cancer risk is crucial for early detection and improved patient outcomes.
• Computed tomography (CT) is a primary imaging modality for lung cancer screening.
• Radiological semantic traits offer potential biomarkers for malignancy prediction.

Purpose of the Study:

• To evaluate the ability of radiological semantic traits from multi-window CT to predict lung cancer risk.
• To determine the added value of combining features from different CT window settings.

Main Methods:

• A cohort of 199 participants (60 lung cancers, 139 benign controls) was analyzed.
• Twenty lung window and 2 mediastinal window features were extracted and scored.
• Multivariate logistic regression and receiver operating characteristic (ROC) analysis were used to assess predictive performance (AUROC, sensitivity, specificity, PPV).

Main Results:

• The combined multi-window CT model significantly improved predictive performance compared to single-window models.
• Area under the ROC curve (AUROC) increased from 0.822 to 0.871 (p=0.009) and 0.877 to 0.917 (p=0.008) at baseline and first follow-up.
• The multi-window model demonstrated superior specificity and positive predictive value (PPV), reaching 0.953 at the second follow-up.

Conclusions:

• Combining semantic features from multiple CT window settings enhances the performance of models for identifying cancerous lung nodules.
• Lung window features were found to be more informative than mediastinal window features for predicting malignancy.