From Pixels to Prognosis: A Comprehensive Review of Classical and Modern Approaches of Lung Nodule Segmentation for Improved Lung Cancer Diagnosis

Lung cancer is a pervasive and life-threatening disease that requires timely detection and treatment for improved patient outcomes. Recent advancements in image processing and deep learning techniques have opened new avenues for identifying cancer in medical images. We examine these studies across various dimensions, encompassing input data (such as data modality, preprocessing techniques, and synthetic data generation), model design (including architecture, modules, and loss functions), and evaluation aspects (covering data annotation requirements and segmentation performance). Our analysis considers mostly the recently proposed methods and adopts a systematic viewpoint to understand the impact of these choices on current trends, and identifies areas (i.e., research gaps), where future researchers can work. To facilitate easy reference and comparison, we have comprehensively summarized the key findings of the existing methodologies. The GitHub repository for this survey paper can be found here.