Development and Validation by deep learning–assisted of predicting growth of persistent pulmonary Ground-Glass nodules
Principal Investigator
Name
Yi Liu
Degrees
M.D., Ph.D.
Institution
Beijing Chao-Yang Hospital
Position Title
Attending physician
Email
About this CDAS Project
Study
NLST
(Learn more about this study)
Project ID
NLST-811
Initial CDAS Request Approval
Jul 15, 2021
Title
Development and Validation by deep learning–assisted of predicting growth of persistent pulmonary Ground-Glass nodules
Summary
The continuous existence of ground glass nodules (GGN) greatly increases the follow-up time and economic cost. Most of those are identified as atypical adenomatous hyperplasia (AAH), adenocarcinoma in situ (AIS), minimally invasive adenocarcinoma (MIA). It is very important to determine the risk factors for GGN to predict the growth of GGNS to shorten the follow-up. In this project, we developed a model to predict the growth of GGN based on chest CT using deep learning.
At present, this project has been approved by the ethics committee, the ethics number is 2021-KE-21. The patient inclusion criteria were as follows: 1) GGN on high resolution pulmonary CT with no change with at least two year follow-up, which was the negative group; 2) GGN advance during the follow-up regardless of the follow-up time. Exclusion criteria: 1) cancer history; 2) interstitial pneumonia, pulmonary sarcoidosis and pneumonia; 3) nodules disappear.
Totally 431 GGN nodules in 214 patients in our center and 100 patients with 130 GGNs of other center have been enrolled, 70% of those are set as the primary group for learning and the rest 30% and nodules in second medical center as the validation group to testify the prediction accuracy. The prediction model established by using U-NET technology. The area under the receiver operator characteristic curve (auROC) validation cohort 1 and validation cohort 2 is 0.81 and 0.75, respectively. However, all the validated data are from China, the reliability of the verification results is questionable. Therefore, external data from different centers is still needed for verification.
Hereby, we apply to use NLST dataset to match those patients with GGNs as the external validation set.
At present, this project has been approved by the ethics committee, the ethics number is 2021-KE-21. The patient inclusion criteria were as follows: 1) GGN on high resolution pulmonary CT with no change with at least two year follow-up, which was the negative group; 2) GGN advance during the follow-up regardless of the follow-up time. Exclusion criteria: 1) cancer history; 2) interstitial pneumonia, pulmonary sarcoidosis and pneumonia; 3) nodules disappear.
Totally 431 GGN nodules in 214 patients in our center and 100 patients with 130 GGNs of other center have been enrolled, 70% of those are set as the primary group for learning and the rest 30% and nodules in second medical center as the validation group to testify the prediction accuracy. The prediction model established by using U-NET technology. The area under the receiver operator characteristic curve (auROC) validation cohort 1 and validation cohort 2 is 0.81 and 0.75, respectively. However, all the validated data are from China, the reliability of the verification results is questionable. Therefore, external data from different centers is still needed for verification.
Hereby, we apply to use NLST dataset to match those patients with GGNs as the external validation set.
Aims
1. Development the GGN growth prediction model based on pulmonary CT using deep learning;
2. Validating the prediction model by internal and external dataset and figure out those GGN with higher risk of growth.
Collaborators
1.Chuang Zhu, Beijing University of Posts and Telecommunications.
2. Minzhen Li,Beijing University of Posts and Telecommunications.
3.Yanhua Tang, Beijing Chaoyang hospital.