Skip to Main Content

An official website of the United States government

Using Low-Dose Lung Computed Tomography to Predict the Risk of Lung Cancer

Principal Investigator
Name
Gigin Lin
Degrees
M.D., Ph.D.
Institution
Chang Gung Memorial Hospital
Position Title
Chair, Department of Radiology
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-764
Initial CDAS Request Approval
Mar 8, 2021
Title
Using Low-Dose Lung Computed Tomography to Predict the Risk of Lung Cancer
Summary
The high incidence and aggressive nature of lung cancer has thwarted efforts to reduce mortality from this cancer through early detection by screening. Data from the National Lung Screening Trial (NLST) from at 33 U.S. medical centers shows that screening with the use of LDCT reduces mortality from lung cancer [1]. The following randomized trial involving high-risk persons, lung-cancer mortality was significantly lower among those who underwent volume CT screening than among those who underwent no screening. There were low rates of follow-up procedures for results suggestive of lung cancer [2]. While both studies are well calibrated at the population level, they are not accurate at the individual level. Therefore, to improve overall outcomes, it is critical to identify cancers earlier; this goal motivates research in creating improved models for cancer risk, and creating personalized screening guidelines based on cancer-risk.

This project is an international collaboration with Professor Regina Barzilay and PhD candidate Adam Yala, Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology (MIT), MA, USA. Recently, MIT developed deep learning models based on full resolution mammograms [3] and found them to be significantly more accurate than the Tyrer-Cusick model, a current clinical standard. Since this publication, MIT has significantly improved the model and extended it to jointly predict the risk of future cancer at one, two, three, four and five years from the mammogram. The model’s ability to predict both near term (within one year), and longer term (within five years) risk will enable new guidelines to personalize the frequency of a woman’s screening based on her individual risk profile.

To continue this endeavor, we would like to combine the NLST and the Chang Gung Memorial Hospital (CGMH) datasets (with different ethnic backgrounds and environmental factors), quantify the difference in performance and study the impact of fine-tuning to improve model generalization. Moreover, we will also quantify the simulated impact of risk-based screening guidelines on the retrospective data. This study is a critical step to move the science forward and bring these tools closer to deployment.
Aims

Study Objectives
Primary
To evaluate the discriminative accuracy of image-based risk models at various time-points (1 to 6 years from LDCT) on study population both with and without fine-tuning the initial weights.
Secondary
To measure simulated impact of risk-based screening guidelines (based on image-based risk models) on the study population in terms total imaging and expected impact on diagnosis time.
Study Design
Duration of Treatment: N/A
Number of Planned Patients
16,000 participants undergoing LDCT screening between the years of 2006-2019, aged 18 years old or above.
For each subject, we aim to collect the following information:
(1) all consecutive LDCT from 2006-2019 and their dates.
(2) the date of the last negative screening follow-up.
(3) date of a pathology confirmed cancer diagnosis (if any).

Collaborators

Professor Regina Barzilay, Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology (MIT), MA, USA.
Adam Yala, PhD candidate , Department of Electrical Engineering and Computer Science, Massachusetts Institute of Technology (MIT), MA, USA.