Novel Integrative Approach for the Early Detection of Lung Cancer using Repeated Measures

This application is being submitted in response to the Notice of Special Interest (NOSI) identified as “NOT-CA24-058.” Lung cancer remains the leading cause of cancer-related deaths worldwide, with early detection being critical for improving prognosis. The National Lung Screening Trial (NLST) provides a comprehensive dataset of longitudinal low-dose CT scans, offering a unique opportunity to study the natural history of lung nodules. This project aims to improve early diagnosis of lung cancer by characterizing the differential trajectory of benign and malignant nodules on serial CTs, identifying and longitudinally tracking all nodules across NLST participants. By modeling the natural course of nodules on serial imaging studies within the NLST, we will expedite the identification of patients who develop cancer and those who do not, and better understand the natural history of individual nodules and nodule loadings within each patient. The parent project, a prospective observational trial, has successfully recruited a diverse cohort of participants and advanced AI-based algorithms for nodule assessment and cancer risk stratification, integrating longitudinal multimodal data. Since our initial proposal in 2019, substantial technological innovations in AI and CT harmonization have emerged, enhancing our potential to accurately characterize lung nodules. Specifically, our team has developed innovative AI-based kernel harmonization techniques and body composition analysis, significantly improving nodule assessment accuracy. These advancements have positioned us well to explore new avenues in lung cancer detection and risk assessment, justifying the need for supplemental funding to integrate these emerging technologies and expand our research scope. The proposed supplemental project introduces innovative approaches to lung cancer detection by longitudinally tracking all nodules across multiple CT scans, providing a dynamic view of nodule behavior. By employing kernel harmonization techniques, we will ensure consistent and reliable biomarker measurements across different machines, acquisition protocols, and reconstructions, enhancing the robustness of our models. Additionally, incorporating radiologic features beyond the spatial boundaries of the nodules, such as tumor-free surrounding lung parenchyma and patient-specific characteristics like body habitus, will offer a personalized understanding of nodule dynamics and lung cancer risk, improving precision in risk stratification and aiding early detection efforts. The overall impact of the project is to improve the early detection and risk stratification of lung cancer by leveraging the comprehensive NLST dataset. By characterizing the natural history of lung nodules and understanding the influence of patient characteristics, the developed models and biomarkers will provide valuable tools for radiologists and clinicians, enhancing precision cancer screening and ultimately reducing lung cancer mortality. The improved understanding of the natural history of lung nodules will ideally position the next stage of efforts for AI validation in randomized control trials, both by our team and others.