Artificial Intelligence & Radiomics for Stratification Of Lung Nodules After Radically Treated Cancer (AI-SONAR)

Improvements in cancer detection and diagnosis have led to increasing numbers of patients being diagnosed with early stage cancer and potentially receiving curative therapy with improved survival outcomes. Recent retrospective studies in cancer survivors have demonstrated such patients possess an increased risk of further cancer in their lifetime compared to the general population, in part potentially due to shared lifestyle risk factors (e.g. smoking), genetic cancer pre-disposition or downstream oncogenic side effects of anti-cancer therapies (eg. radiotherapy). Lung cancer remains the leading cause of cancer related deaths worldwide and the lungs also represent a common site for metastatic disease in patients with non-pulmonary malignancy. Furthermore, lung cancer is one of the most common second primary malignancy in patients with a prior history of treated cancer. Therefore, discerning the significance of a pulmonary nodule in the context of a previous cancer remains a clinical challenge given it may possess the potential to represent benign disease, metastatic relapse or new primary malignancy.

This study will assess the utility of radiomics and artificial intelligence approaches to new lung nodules in patients who have undergone radical treatment for a previous cancer. This will entail use of machine learning (ML) approaches and later, exploration of deep-learning/convolutional neural network approaches to nodule interpretation for differentiation of benign, metastatic and new primary lung cancer nodules/lesions. Development of a ML classifier or deep learning based tool may help guide which patients would benefit from earlier investigations including additional imaging, biopsy sampling and lead to earlier cancer diagnosis, leading to better patient outcomes in this unique cohort. This is a retrospective study analysing data already collected routinely as part of patient care. All data will be anonymised prior to any analysis, no patient directed/related interventions will be employed and consent-waiver for study inclusion will be exercised.