Development of radiomic models for lung nodule diagnosis
This project will analyze the NLST dataset of low-dose CT scans, including scans with both benign and malignant nodules. Features will be developed and extracted from left and right lung fields of all patients and algorithms will be developed and extracted to analyze the association with benign/malignant status. These analyses will be able to quantitatively describe the natural range of all features as well as their reproducibility. The impact of these studies will be the identification of a panel of features that can characterize patients who have benign nodules, and reduce the need for further more invasive diagnostic procedures.
Aim 1. Develop robust methods to segment both the lung fields of normal patients and also patients with lung nodules. We will use our newly developed artificial segmentation program.
Aim 2. Identify an NLST low-dose CT dataset sample that will be representative of the entire set. To avoid mining of unreliable data, we will need to include all scans of patients with confirmed malignant lung nodules and select a benign sample that is well-matched. This data sample will be used to validate our feature extraction software and radiomics model.
Aim 3. Extract and analyze data from the NLST dataset sample. Features will be extracted from all validated patients in the NLST dataset sample for both L and R lung fields in all three longitudinal scans from each participant. From this data, unequivocally negative/benign nodules and these will be used to develop a baseline normal set of features to represent benign features. Likewise, unequivocally malignant nodules will also be extracted and analyzed to compare with the baseline set and identify distinguishing features which are highly stable, and thus reproducible.
Shawn Sun, Columbia University Medical Center
Lin Lu, Columbia University Medical Center
Hao Yang, Columbia University Medical Center
Bingsheng Zhao, Columbia University Medical Center