Study
NLST
(Learn more about this study)
Project ID
NLST-942
Initial CDAS Request Approval
Aug 2, 2022
Title
A semi-supervised deep clustering framework for personalized post-test risk-stratification in lung cancer screening
Summary
The efficacy of low-dose computed tomography (LDCT) screening in decreasing lung cancer (LC) mortality has been consistently demonstrated, and intensive investigation is ongoing, aiming to better profiling of subjects to be invited and to make screening more cost-effective. In particular, the estimated risks can also include post-test radiological variables related to other smoking-associated diseases such as chronic obstructive pulmonary disease (COPD) and cardiovascular disease (CVD), which may decrease the efficacy of LDCT screening since they are cause-of-death strong competitors. Accordingly, outcomes for personalized risk-stratification may be extended from LC incidence to mortality due to LC, respiratory disease, and CVD with implications on whom to screen, how frequently, and until when. The variable combination, in a single subject, of the smoking-related comorbidities, namely LC, COPD, and CVD, can be investigated by identifying distinct diseases through a semi-supervised deep learning paradigm. According to this hypothesis, in this project, we propose a deep learning framework based on GANs (Generative Adversarial Network) tailored to examine risk factors for LC and radiological features of COPD, and CVD in a cohort of smokers and former smokers undergoing LC screening with LDCT. Moreover, using longitudinal data, we may identify distinct evolution pathways for LC, COPD, and CVD diseases, including death from these smoking-related conditions. We will be able to exploit the full possibility of this dataset for predicting short-term (LC incidence) and long-term outcomes (mortality due to LC, respiratory disease, and CVD) at different times after baseline LDCT.
Aims
O1) Development of a semi-supervised deep clustering framework for personalized risk-stratification in lung cancer screening
O2) Internal validation of the semi-supervised deep clustering framework models for multi-level risk-stratification in LC screening tailored to single subjects. We will develop and internally validate multi-level post-test risk-stratification model combining information gathered from the subject's age, pack-years, and additional historical information, and enriched post-test LDCT models. We shall add to the baseline LDCT data the severity of pulmonary emphysema assessed with densitometry and Coronary Artery Calcifications using a 0-4 scores comprehensive visual scale to stratify the risk of LC incidence. Moreover, we shall calculate the prediction value of the models for mortality from LC, respiratory disease, and cardiovascular disease.
Collaborators
Giulia Raffaella De Luca, Department of Electrical, Electronic, and Information Engineering "Guglielmo Marconi", University of Bologna
Dr. Matteo Lai, Department of Electrical, Electronic, and Information Engineering "Guglielmo Marconi", University of Bologna
Dr. Chiara Marzi, Institute of Applied Physics “Nello Carrara” – IFAC, National Research Council – CNR
Prof. Mario Mascalchi, "Mario Serio" Department of Experimental and Clinical Biomedical Sciences, University of Florence
Dr. Riccardo Scheda, Department of Electrical, Electronic, and Information Engineering "Guglielmo Marconi", University of Bologna
