Image mining and ctDNA for improved risk stratification and outcome prediction in NSCLC by application of artificial intelligence

Lung cancer presents a major global health challenge, necessitating advancements in diagnostic and treatment approaches [1]. The integration of cutting-edge technologies, particularly Artificial Intelligence (AI), has become crucial for enhancing diagnostic accuracy and treatment efficacy in the complex landscape of lung cancer. AI applications in diagnostic imaging, including computed tomography (CT) and positron emission tomography (PET), demonstrate remarkable capabilities in improving the precision and efficiency of cancer diagnosis. Through machine learning (ML) and deep learning (DL), AI systems are able to not only detect subtle abnormalities, but also extract complex patterns and features that may elude human perception [2,3]. Additionally, the advent of liquid biopsy, particularly the analysis of circulating tumour DNA (ctDNA), has paved the way for non-invasive cancer diagnosis and monitoring [4]. Integrating AI into diagnostic imaging analysis and interpretation of liquid biopsy data could potentially improve lung cancer staging and the prediction of treatment response, disease recurrence and overall patient prognosis. This thesis explores the synergistic potential of AI, diagnostic imaging, and liquid biopsy as a multidimensional approach to unravelling the complex landscape of lung cancer. Chapter 1 serves as a broad introduction and aims to provide a comprehensive discussion of the collaborative impact of these disciplines in significantly advancing both diagnosis and treatment strategies for lung cancer. Chapter 2 focuses on the preliminary results of the main project, which integrates AI, liquid biopsy, and imaging to predict patient staging and outcomes. Chapter 3 examines the controversies of using semi-quantitative PET measures, such as SUVmax, as diagnostic and prognostic biomarkers in patients with suspicious lung lesions. Finally, Chapter 4 explores the application of triplet networks, addressing the challenges associated with unbalanced and limited data sets in lung cancer applications. Chapter 5 provides a comprehensive discussion of the key findings from the three studies included in the thesis.