Cancer Risk Estimation Combining Lung Screening CT with Clinical Data Elements.

Area of Science:

• Radiology and Medical Imaging
• Oncology
• Artificial Intelligence in Healthcare

Background:

• Lung cancer risk prediction is crucial for early detection and intervention.
• Existing models often rely solely on clinical data elements (CDEs) or imaging data, limiting their predictive power.
• Manual interpretation of CT scans for risk assessment is time-consuming and requires specialized expertise.

Purpose of the Study:

• To develop an automated model for lung cancer risk estimation.
• To integrate chest CT scan data with CDEs for enhanced prediction accuracy.
• To eliminate the need for manual reading of CT scans in risk assessment.

Main Methods:

• Retrospective study of two screening cohorts: National Lung Screening Trial (NLST) and Vanderbilt Lung Screening Program (VLSP).
• Development of a co-learning model utilizing whole CT scans and CDEs, assessed via fivefold cross-validation on the NLST dataset.
• External validation of the model on the VLSP dataset, comparing its performance against CDE-only, image-only, and Brock models using AUC and AUPRC metrics.

Main Results:

• The co-learning model achieved a higher Area Under the Receiver Operating Characteristic Curve (AUC) of 0.88 on the NLST dataset compared to CDE-only (0.69) and image-only (0.86) models (P < .05).
• On the external VLSP dataset, the co-learning model demonstrated superior performance with an AUC of 0.91, significantly outperforming CDE-only (0.59) and image-only (0.88) models (P < .05).
• The proposed model also outperformed the Brock model in lung cancer risk prediction.

Conclusions:

• Combining chest CT images and CDEs in a co-learning model significantly enhances lung cancer risk prediction accuracy.
• The developed model offers a more effective alternative to prediction models relying solely on CDEs or imaging data.
• This automated approach holds promise for improving lung cancer screening efficiency and patient outcomes.