Deep Learning Approaches for Lung Cancer Detection from CT Images: A Review of Challenges and Emerging Research Perspectives

Lung cancer is a predominant universal illness and is an underlying cause of a number of cancer deaths annually. One of the most effective solutions is early detection which may significantly improve treatment and lower mortality, though not as easy as the number of medical imaging data increases continuously and experienced radiologists are scarce. Such impediments (among others) imply that in most rural and resource constrained healthcare facilities, access to advanced investigative facilities is inadequate. CT imaging has been deemed as a potent tool that can be used to identify lung cancer that correlates pulmonary nodule and manual analysis of the CT scans is very tedious and can be characterized by clinical interpretation variability. The current state of research in algorithms in the field of artificial intelligence, deep learning, proved itself to be highly promising in assisting the analysis of medical images and clinical decision-making. These models, however, are based on complicated structures where they demand the utilization of high-performance computing devices and substantial memory space that makes it unfeasible to be utilized daily in clinical practice and edge-based healthcare services. Further, the works that concentrate on advancing accuracy are numerous compared to the effort being put on the efficiency of the models that can be deployed in practice.

In this review, the current research trends in automatic lung cancer detection are discussed, and the major weaknesses of the current approaches limiting its practical implementation are identified. This study identifies the main issues that demand more universal, lightweight and effective solutions, for early diagnosis applications when access to such intelligent health care systems is most crucial.