Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Qian Tan
Degrees
master
Institution
South China University of Technology
Position Title
Graduate Student
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-1343
Initial CDAS Request Approval
Oct 21, 2024
Title
Deep Learning-Based Coronary Calcification Risk Prediction: A Clinical Application Study Using the NLST Dataset
Summary
This study aims to develop and validate a deep learning model based on CT imaging data from the NLST (National Lung Screening Trial) dataset to predict coronary artery calcium (CAC) scores and stratify cardiovascular disease risk. The goal of this project is to create an innovative non-invasive prediction tool by combining CT imaging data from the NLST dataset with other relevant clinical information. This tool will help improve the early diagnosis of coronary artery disease, reduce the need for invasive testing, and provide more reliable evidence for the prevention and treatment of cardiovascular diseases.
We chose to use the NLST dataset because it is large in scale, contains rich CT imaging data, and has comprehensive clinical information related to cardiopulmonary health, which can support our diverse research needs. Additionally, the NLST's multi-center design and strict follow-up standards will ensure the broad applicability and accuracy of our research results. Ultimately, this study aims to optimize the detection of coronary artery calcium through deep learning algorithms and promote the development of low-cost, non-invasive cardiovascular disease risk prediction tools.
Aims

Aim 1: Develop a deep learning model to predict Coronary Artery Calcium (CAC) scores using NLST CT images.

We aim to build and optimize a deep learning model that can predict the presence and extent of coronary artery calcification (CAC) from CT images in the NLST dataset. This model will be trained to assess CAC scores, an established indicator of atherosclerotic cardiovascular disease risk, providing a non-invasive tool for cardiovascular risk stratification.
Aim 2: Validate the deep learning model using NLST clinical data.

To ensure accuracy and clinical relevance, the developed model will be validated against existing clinical records and follow-up outcomes from the NLST cohort. This validation will assess how well the model predicts CAC scores compared to standard manual interpretations and evaluate its potential for integration into clinical workflows.
Aim 3: Stratify cardiovascular disease risk in NLST participants using predicted CAC scores.

Using the predicted CAC scores from the deep learning model, we will stratify participants into low, intermediate, and high cardiovascular disease (CVD) risk categories. This stratification will be compared with traditional risk prediction methods, such as the Framingham Risk Score, to evaluate the model’s incremental value in predicting adverse cardiovascular events.
Aim 4: Investigate the added value of incorporating other clinical and demographic variables to improve prediction accuracy.

We will explore the potential of enhancing the model’s prediction accuracy by incorporating additional variables, such as age, smoking status, family history of cardiovascular disease, and other relevant clinical factors available in the NLST dataset. This aim will evaluate whether combining multiple risk factors improves the model’s performance and utility in risk prediction.
Aim 5: Assess the model’s potential for broad clinical implementation and cost-effectiveness.

We will conduct a preliminary cost-effectiveness analysis to determine whether using this deep learning-based approach to predict CAC scores from routine CT scans provides a more affordable and accessible alternative to current methods. We will also investigate its scalability and applicability in other clinical settings for early cardiovascular risk detection.

Collaborators

Yling Tao