A deep-learning signature of lesion detection, malignancy evaluation, risk stratification and management strategy for pulmonary nodules on computed tomography
Two datasets, the NLST dataset based on the screening population and the surgery dataset from our center based on the treatment population, will be utilized for the model construction to potentiate the generalization and robustness of the deep-learning signature in the clinical scenarios. Furthermore, by comparing to two conventional clinical models for determining malignancy risk of pulmonary nodules (Lung-RADS Model and Brock Model), the diagnostic efficacy of the deep-learning signature will be further validated. Subsequently, we will investigate the value of the malignancy score generated from the deep-learning signature in prognosis stratification of pulmonary nodules. Still, on the basis of three-year follow-up CT images, a reliable growth model will be established to optimize the management strategy of pulmonary nodules.
In summary, our proposed deep-learning signature will provide instructive significance in individualized malignancy evaluation, prognosis stratification and management strategy decision-making for patients with indeterminate pulmonary nodules.
Aim #1: To develop a robust deep-learning signature of lesion detection and malignancy evaluation for pulmonary nodules.
Aim #2: To indicate the malignancy extent of the lung nodules based on the generated risk score from deep-learning signature and thus provide prognosis stratification for patients with indeterminate pulmonary nodules.
Aim #3: To establish a growth model based on three-year follow-up CT images and provide instructions for the optimal management strategy of pulmonary nodules.
Chang Chen, M.D., Ph.D., Shanghai Pulmonary Hospital, Tongji University School of Medicine
Jiajun Deng, M.D., Shanghai Pulmonary Hospital, Tongji University School of Medicine
Yifan Zhong, M.D., Shanghai Pulmonary Hospital, Tongji University School of Medicine
Jiancheng Yang, Ph.D., Shanghai Jiao Tong University
Shouyu Cheng, Ph.D., Tongji University