Early Lung Nodule Malignancy Assessment Using Longitudinal Chest CT Data

Principal Investigator

Name
Toru Higaki

Degrees
Ph.D.

Institution
Hiroshima University

Position Title
Associate Professor

Email
higaki@hiroshima-u.ac.jp

About this CDAS Project

Study
NLST (Learn more about this study)

Project ID
NLST-1534

Initial CDAS Request Approval
Oct 12, 2026

Title
Early Lung Nodule Malignancy Assessment Using Longitudinal Chest CT Data

Summary
This study aims to develop a machine learning–based framework for assessing the malignancy of pulmonary nodules using chest CT imaging data.

Longitudinal CT scans acquired from the same participants over multiple screening years will be utilized during model training to learn imaging characteristics associated with malignant and benign nodules. Rather than relying on explicit temporal progression at inference time, this study focuses on leveraging longitudinal information during training to capture imaging features that may already be present at early screening time points.

Chest CT images from the National Lung Screening Trial (NLST), together with associated clinical outcome data (e.g., lung cancer diagnosis status and diagnosis year), will be used for model development and evaluation. Clinical data will be utilized solely for outcome labeling and alignment of imaging data across screening rounds.

All analyses will be conducted using de-identified data, and the data will not be used for clinical decision-making. The results will be reported only in aggregate form.

Aims

• To construct a longitudinal chest CT dataset by aligning serial scans acquired from the same participants across multiple screening rounds and linking these imaging data with lung cancer outcome information, including diagnosis status and diagnosis year.

• To develop a machine learning model that leverages longitudinal imaging data during training to learn imaging characteristics associated with pulmonary nodule malignancy, while allowing flexibility in model design and feature representation.

• To investigate whether imaging features learned from longitudinal data enable accurate malignancy assessment when using early or limited CT scans, such as baseline or initial screening images, and to evaluate the potential of such features for early-stage malignancy characterization in a research setting.

Collaborators

Toru Higaki Hiroshima University
KOSUKE KOBAYASHI Hiroshima University