Photo Credit: David Gyung
At the 2023 North American Conference on Lung Cancer, held December 1-3 in Chicago, Florian Fintelman, MD, presented as part of the “Strategies for Early Detection, Screening, Risk Reduction, and Cure” session, a recap of data on the use of artificial intelligence (AI) for lung cancer risk prediction on chest CT.
Dr. Fintelman had two recently published papers that address this topic, which we summarize here:
Role of Sex in Lung Cancer Risk Prediction Based on Single Low-Dose Chest Computed Tomography
Sci Rep. 2023 Oct 30;13(1):18611.
Summary: The open-source deep-learning algorithm Sybil, validated for predicting long-term lung cancer risk from low-dose chest computed tomography (LDCT), exhibits high accuracy in a real-world context. Trained on a predominantly male cohort, concerns arise about potential gender-based discrepancies. Analyzing 10,573 LDCTs from 6,127 lung cancer screening participants over six years, Sybil demonstrated comparable predictive performance for males and females at the one-year mark. However, at six years, the algorithm displayed superior accuracy in predicting lung cancer risk for females, with an AUC of 0.87, compared to males, where the AUC was 0.79. This gender-based variation suggests potential inequities in long-term risk prediction, emphasizing the importance of addressing imbalances in training data when developing artificial intelligence algorithms for healthcare applications. Despite these disparities, Sybil remains a valuable tool for assessing lung cancer risk, particularly excelling in predicting long-term outcomes for females in a real-world setting.
Artificial Intelligence and Machine Learning in Lung Cancer Screening
Thorac Surg Clin. 2023 Nov;33(4):401-409.
Summary: Artificial Intelligence and Machine Learning (AI/ML) advancements present significant potential in overcoming challenges in lung cancer screening and enhancing health equity. This review explores the current status and future prospects of AI/ML tools within the lung cancer screening workflow. The focus areas include determining eligibility for screening, reducing radiation doses, and image denoising in low-dose chest computed tomography (CT). Additionally, AI/ML plays a crucial role in lung nodule detection, classification, and establishing optimal screening intervals. These tools prove effective in assessing chronic diseases through CT, opening avenues for opportunistic screening and thereby contributing to the improvement of population health. The integration of AI/ML in lung cancer screening not only addresses current obstacles but also holds promise for advancing healthcare practices and promoting inclusivity in health outcomes.