Skip to Main Content

An official website of the United States government

Principal Investigator
Name
Yi Xu
Degrees
Ph.D.
Institution
Shanghai Jiao Tong University
Position Title
Associate Professor
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-517
Initial CDAS Request Approval
May 31, 2019
Title
Pulmonary nodule diagnosis, follow-up recommendation and lung cancer prognosis
Summary
Lung cancer is the deadliest type of cancer worldwide, posing a serious threat to human life. Early lung cancer is difficult to screen due to its fewer clinical symptoms, so lung nodule detection and diagnosis based on CT image analysis has become a crucial method of early diagnosis of lung cancer. However, lung cancer screening via CT images is a heavy, repetitive thus error-prone task for radiologists. So we hope to develop a nodule detection and diagnosis model leveraging the large amount of annotated NLST CT images at first and get a reliable diagnosis result.
In the second step, if a nodule is detected but not diagnosed with cancer, we will give follow-up recommendations according to its specific growth stages (volume, density, image features, etc.). To achieve this, we will explore a nodule growth model based on three-year follow-up CT images. On the other hand, if a nodule is diagnosed with cancer, we will utilize CT images and patient characteristics to develop a prognosis prediction model. A relatively accurate survival prediction will be given and we hope to improve treatment decision via this model.
In brief, early lung cancer screening, pulmonary nodule follow-up and lung cancer prognosis have significant clinical meanings. Our deep learning-based study will follow such clinical process to lighten the workload of radiologists and reduce the mortality of lung cancer.
Aims

Aim #1: To develop a robust pulmonary nodule diagnosis model (including nodule detection, benign/malignant classification, etc.) with high performance.
Aim #2: For nodules detected but not diagnosed with cancer by model in aim #1, we aim to find an appropriate growth model based on three-year follow-up CT images and give follow-up recommendations according to the specific growth stages.
Aim #3: For nodules diagnosed with cancer (malignant), we aim to develop a novel prognosis prediction model leveraging CT images and patient characteristics.

Collaborators

Tiancheng Lin, Ph.D., SJTU
Yamin Li, M.D., SJTU
Dayi Li, M.D., SJTU