Deep based cancer/nodule detection project
Principal Investigator
Name
WenJia Song
Degrees
M.D.
Institution
University of Queensland
Position Title
PhD student
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-894
Initial CDAS Request Approval
Mar 22, 2022
Title
Deep based cancer/nodule detection project
Summary
This project aims to develop a deep learning (DL) method for lung cancer screening with CT scans. We wish to build two deep learning models based on the dataset. The first one is the nodule detection model, which takes the whole CT scan as the input and will try to find the region of interest. The position and diameter of the nodule will work as a label or ground truth. Once the training step is completed, the module will be able to predict the possible position for the nodule. The second model will be used to detect the key information, which may be able to predict the required information including the cancer probability. This model will take one or several CT scans of one patient as input, and the relationship between the details of the CT scan and the required feature will be learned. In addition, 70 percent of data will be used for training, while 20 percent of data will be used for validation. The rest of the data will be used for testing.
Aims
We plan to build a deep learning model based on an end-to-end pipeline that takes the whole CT scan as input for lung cancer prediction. It will involve several steps: (1) The model will automatically detect the ROI (Region of interest), and the model will then be able to tell whether the ROI is a nodule or not. (2) The ROI, which has been determined as a nodule, will be considered as an input for the second model. Taking the ROI and the whole CT scan as the input, the 2nd model will predict the required information (e.g., Cancer probability…). However, additional labels and data may be required for this stage.
Collaborators
Prof. Feng Liu The University of Queensland
Dr. Fangfang Tang The University of Queensland