Deep learning predicts cardiovascular disease risks from lung cancer screening low dose computed tomography

Area of Science:

• Radiology
• Cardiology
• Artificial Intelligence

Background:

• Cancer patients face elevated cardiovascular disease (CVD) mortality risks.
• Low-dose computed tomography (LDCT) screening for lung cancer presents a potential avenue for concurrent CVD risk assessment.

Purpose of the Study:

• To develop and validate a deep learning model for predicting CVD mortality risk using LDCT scans.
• To assess the efficacy of LDCT as a dual-purpose screening tool for both lung cancer and CVD risk.

Main Methods:

• A deep learning model was trained on 30,286 LDCT scans from the National Lung Cancer Screening Trial.
• Model performance was evaluated using area under the curve (AUC) on separate test sets.
• Validation was performed against established cardiac CT markers like coronary artery calcification (CAC) and MESA 10-year risk scores using an independent dataset.

Main Results:

• The deep learning model achieved an AUC of 0.871 for CVD risk prediction on a test set.
• The model demonstrated an AUC of 0.768 for identifying patients with high CVD mortality risk.
• Validation confirmed the model's ability to predict CVD risk comparable to traditional cardiac CT markers.

Conclusions:

• Deep learning models can effectively utilize LDCT scans for simultaneous lung cancer and CVD risk screening.
• LDCT can be repurposed as a quantitative tool for dual-risk assessment in high-risk populations.
• This approach offers a more efficient screening strategy for patients at risk of both lung cancer and cardiovascular disease.