https://biometry.nci.nih.gov/cdas/approved-projects/1853/

Improving Clinical Classification of a patient’s lung tumors, NSCLC TNM cancer stage

Principal Investigator

Name

Dolly Wu
Degrees
Ph.D., J.D.
Institution
UT Southwestern Medical Center
Position Title
Instructor
Email
dolly.wu@utsouthwestern.edu

Project Information

Study

NLST
Project ID
NLST-416
Title
Improving Clinical Classification of a patient’s lung tumors, NSCLC TNM cancer stage
Summary
Improve the accuracy and consistency of the clinical TNM classification for NSCLC cancer by implementing machine algorithms. The most-used staging system is the TNM (primary Tumor size, affected nearby lymph Node, distant Metastasis) cancer classification. T, N, and M will be classified together by software, which requires new algorithms that use both imaging and non-imaging data (e.g. medical tests), and new methods to generate training data.

Aims

Aim 1: Use CT image data to train neural network algorithm. Also create simulated lung cancer database from NCI’s NLST (National Lung Screening Test) non-contrast CT database, add simulated tumors in the CT database. The simulated dataset is used to train deep neural networks in Aims 2 and 3. Would like to get the CT scans of participants who have NSCLC and also those who do not.
Aim 2: Determine the clinical T (size of a primary tumor) because the classes of T changed much in 2018 and the tumor size measurement can be subjective. Also, we can optimize the 3-variable TNM network architecture by classifying one variable, T, alone.
Aim 3: Assess best ways to include non-image data with image data, and estimate results. We use simulated image data to train our deep network to classify TNM. We will test our trained network on real patient data to evaluate the best approach, and thereby also get a preliminary estimate of the TNM classification accuracy.
Aim 4: Also like to get the full set of CT scans of participants who actually had NSCLC, to try training/testing the neural net. And to verify the accuracy of the simulated dataset.
Aim 5: Analyze correlations among different variables.

Collaborators

Kiran Batra, MD, UTSW
Sharon Hoskin, RN, UTSW
Nhat-Long Pham, MD, UTSW
Stephen Seiler, MD, UTSW
Ann Spangler, MD, UTSW
Jing Wang, PhD, UTSW
Yulong Yan, PhD, UTSW