Skip to Main Content

An official website of the United States government

Early Detection and Prognosis of Lung Cancer

Principal Investigator
Name
Yun Li
Degrees
Ph.D.
Institution
UNC Chapel Hill
Position Title
Associate Professor
Email
About this CDAS Project
Study
NLST (Learn more about this study)
Project ID
NLST-262
Initial CDAS Request Approval
Dec 2, 2016
Title
Early Detection and Prognosis of Lung Cancer
Summary
We focus on both imaging data and genetic data from NLST on early tumor detection and prognosis of lung cancer. Traditionally, early detection performed by medical experts is time consuming and subtle image details may not be visually discernible by human eyes, which contributes to the inaccuracy of early tumor detection. The identification of key features and automation of the detection procedure is challenging due to the small signal-noise ratio, as well as the large variation of the volume, location, and other characteristics of tumors across individuals. Machine learning techniques will be considered to automate and standardize the procedure of early tumor detection, which will increase accuracy and efficiency in both time and cost. An automated early detection approach that is more accurate and cost-efficient will assist the spread of cancer screening, which can possibly be generalized to other cancer types. After the location of the tumor is detected, the tumor progression will be further monitored, based on which the possibility of whether the tumor is benign or malignant will be predicted. We can then classify the cancer stage, improve the treatment decision, and hence increase patient survival.
Aims

Our two major aims are as follows:

1. The first aim is early detection of lung cancer incorporating the use of genetic marker in cancer screening. For patients with existing information on their genetic traits or markers, this information can be utilized to update their likelihood of having the cancer and increase the accuracy of early tumor detection. We will also combine methods on early tumor location detection by extracting the medical image feature and the established uniform coordinate system using image registration technique. Different voxel-based machine learning techniques will also be used to segment the active tumor from normal tissue , necrosis, and surrounding edema area.

2. The second aim is tumor Prognosis prediction, based on both imaging data and genetic traits when monitoring the tumor progression using longitudinal datasets and identify key features related to tumor metastasis. Statistical and computational methods including deep learning, neural networks, and random forests will be investigated. Targeted treatments identify and attack cancer cells with specially designed substances, while doing as little damage as possible to normal cells so as to minimize the side effect and healthcare costs. A reliable prediction can enhance the targeted treatment.

Collaborators

Dr. Hongtu Zhu, UNC Chapel Hill
Yue Shan, UNC Chapel Hill