Enhancing Early Lung Cancer Diagnosis: Predicting Lung Nodule Progression in Follow-Up Low-Dose CT Scan with Deep...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Oncology

Background:

• Early lung cancer diagnosis is crucial for improving patient survival rates.
• Current follow-up protocols for lung nodules involve periodic low-dose computed tomography (LDCT) scans.
• Predicting nodule growth can aid in timely diagnosis and treatment decisions.

Purpose of the Study:

• To develop and evaluate a Growth Predictive model based on Wasserstein Generative Adversarial Network (GP-WGAN) for predicting lung nodule growth patterns.
• To assess the performance of predicted nodule images (GP-nodules) in lung cancer diagnosis.
• To compare the GP-WGAN model's performance against existing methods like Lung-RADS and the Brock model.

Main Methods:

• A GP-WGAN was trained on 1121 pairs of nodule images from LDCT scans with approximately 1-year intervals.
• The model predicted 1-year follow-up nodule images (GP-nodules) for 450 nodules in an independent test set.
• A lung cancer risk prediction (LCRP) model classified both GP-nodules and real follow-up nodules.

Main Results:

• The LCRP model achieved an AUC of 0.827 ± 0.028 when classifying GP-nodules.
• Performance using GP-nodules was comparable to using real follow-up nodule images (AUC = 0.862 ± 0.028, p = 0.071).
• The GP-WGAN model significantly outperformed Lung-RADS and the Brock model (p < 0.05).

Conclusions:

• Predicted lung nodule growth using GP-WGAN shows comparable diagnostic performance to real follow-up scans.
• This AI-driven approach holds potential for earlier lung cancer detection and intervention.
• Integrating GP-WGAN into clinical practice could accelerate diagnosis over traditional waiting periods.