LungHealth AI: Using AI to improve the early detection and diagnosis of lung cancer

Identification of patients with non-specific but concerning symptoms for cancer (NSCSs) for further investigation is challenging. Approximately 50% of cancer patients present with NSCSs [1]. They are typically referred later to cancer services and have poorer outcomes [2]. We will be taking an initial focus on lung cancer, which is the most frequent cause of cancer-related deaths worldwide. According to CRUK, the net five-year survival rate is 13.8% [3] and 70% of lung cancers are detected at later stages of the disease, when it is harder to treat. Early diagnosis of lung cancer may allow more treatment options and therefore improve patient survival. Lung abnormalities cannot be seen or felt by the patient, and thus consultations are usually triggered when later NSCSs appear, such as a persistent cough, chest pain, shortness of breath or unexplained weight loss. However, these do not always lead to further investigations for lung cancer, especially in non-smokers; contributing to many cancers only being detected on emergency admission following additional symptoms. Recent evidence demonstrates that low-dose CT screening for lung cancer can facilitate early detection and reduce mortality [4-6]. However, several challenges remain including the unmet need of how best to select subjects to screen in a cost-effective manner [7, 8]. A problem shared with current referral from primary care is the exclusion of non-smokers, in which 15-20% of lung cancers occur [9]. Non-smokers, especially younger ones, are rarely referred for diagnosis, because they are not considered at risk. Even for smokers, the nature of NSCSs can significantly delay their referral. However, they could potentially help in triaging patients better, together with other factors, many routinely collected, such as, but not limited to, demographic, lifestyle, pre-existing medical conditions, diagnosis data, hospital episodes statistics (HES). LungHealth AI is an ambitious, ground-breaking, disruptive artificial intelligence (AI) system aiming to transform the early diagnosis of lung cancer, based on a holistic approach, powered by interdisciplinarity (computational methods, AI and statistics along with clinical guidance), using longitudinal data to discover the hidden spatiotemporal relationships and/or patterns within multimodal data and their evolution over the time. Finding such evolving patterns is paramount to plausible early detection and timely interventions to address the challenges of lung cancer prevention and targeted treatment. We hope to identify the patterns providing predictors of lung cancer in a non-invasive manner for efficient diagnosis referral or improved inclusion in screening, resulting in earlier detection. It would also help identify candidates for preventative treatments. Successful results from this pilot project will validate the approach for further investigation in larger, population-based datasets. LungHealth AI would potentially a steppingstone to transform the early detection and diagnosis of lung cancer, resulting in significant health, social and economic benefits to the UK, and beyond.