A Review of Advances in Computational Predictive Modeling for Lung Cancer Diagnosis

Abstract

Lung cancer is one of the world’s leading causes of cancer-related death due to the challenges associated with late-stage diagnosis and the difficulty associated with accurately identifying malignant lung lesions in a timely manner. As a result, there has been extensive development of various computational predictive modeling approaches that employ the latest advancements in computational and machine learning techniques (including deep learning, radiomics, etc.), as well as the use of multiple-modality data sources for accurate diagnosis of lung cancer via medical imaging, clinical data and biological data. This review provides a summary of recent research and developments surrounding computational predictive modeling of lung cancer diagnosis, specifically emphasizing the use of deep learning models, transformer networks and multi-modal frameworks due to their unique ability to automatically extract complex features, provide context globally and integrate multiple-source data for improved diagnostic accuracy. A systematic synthesis of the currently available methodologies, datasets and evaluation strategies will be presented by analyzing the advantages these approaches have in regard to being accurately scalable, transferable to the clinical setting, and applicable clinically. This review aims to provide new information concerning current strengths and weaknesses while providing insight into the most useful approaches available at this time for the early detection of lung cancer, thus assisting with better decision-making in clinical settings.