AI for Lung Cancer Risk Definition in Computed Tomography Screening Programs

Overview

Low-dose computed tomography (LDCT) lung cancer (LC) screening can reduce mortality among heavy smokers, but there is a critical need to better identify people at higher risk and to reduce harms related to management of benign nodules. The most promising strategy is to combine novel tools to optimize clinical decisions and increase the benefit of screening. In this respect, the investigators already demonstrated that the combination of baseline LDCT features with a minimal invasive microRNA blood test was able to more precisely estimate the individual risk of developing LC. The investigators posit that additional immune-related and radiologic features can be integrated with the help of artificial intelligence (AI) to further implement LDCT screening strategies. The project will answer whether the combination of (bio)markers of different origin can predict LC development at baseline and over time, indicate which screen-detected lung nodules are likely to be malignant and ultimately reduce LC and all cause mortality.

Lung cancer constitutes 28% of all cancer deaths in Europe, with 70% of patients diagnosed at advanced stages and a mere 21% 5-year survival rate. Despite smoking's causative link to almost 90% of cases, global smoking rates persist, posing a long-term public health challenge. Our focus lies in refining lung cancer risk assessment using blood-based biomarkers, particularly circulating microRNAs (miRNAs) and C-reactive protein. Biennial LDCT screenings and blood tests predicting lung cancer risk have shown effectiveness, as seen in our pioneering work within the BioMILD trial since 2013. The BioMILD trial, encompassing 4119 volunteers, combines LDCT and microRNA biomarkers, demonstrating feasibility and safety over 4 years. Our current endeavor aims to develop a predictive model for LDCT-detected high-risk lung nodules, incorporating blood, functional, and radiomics biomarkers. Leveraging the BioMILD trial's biorepository, imaging database, and 20 patient-derived xenografts (PDXs), the investigators utilize advanced artificial intelligence (AI) tools for comprehensive analysis. This approach, involving 400 subjects with solid and sub-solid LDCT lung nodules, including 100 baseline-identified cancer patients, is crucial. By combining blood-based biomarkers, radiologic parameters, clinical features, and AI tools, the investigators aim to create a robust model. This model will be validated using an independent set of 100 subjects (25 with and 75 without lung cancer) from the ongoing SMILE screening trial. If successful, our vision is to prospectively implement this panel in clinical contexts where it proves beneficial. Our mission is to reduce lung cancer mortality, optimizing screening interventions with novel, non-invasive tools for all high-risk individuals while minimizing costs and radiation exposure-related harms. Aim 1 Assessment of an Immune Signature Classifier (ISC) on peripheral blood mononuclear cell (PBMC) samples collected from screen detected solid and sub-solid LDCT lung nodules and integration of ISC with existing biomarkers such as the MSC test and the c-Reactive Protein (cRP). Aim 2 Evaluation of radiologic features and other LDCT markers related to respiratory and cardiovascular disorders. Aim 3 Development of a risk classifier using AI tools based on combination of blood biomarkers, imaging and clinical data to improve LDCT screening sensitivity and positive predictive value.