Title : Classification of lung cancer as stage I or stage II using computer vision and deep learning techniques
Abstract:
Lung cancer is still the primary cause of cancer deaths globally, and precise discrimination between Stage I and Stage II disease is important to inform appropriate treatment planning. Manual staging by computed tomography (CT) continues to be time-consuming, subject andsusceptible to inter-reader variability. To address this, the current review provides an extensive overview of recent deep learning-based methods solving lung nodule segmentation,feature extraction, and automated staging from CT scans. We integrate literature on threeprimary themes: (1) segmentation algorithms—particularly U-Net and its extensions for lungand nodule segmentation; (2) classification models using convolutional neural networks (e.g.ResNet, VGG16, MobileNet) and radiomics features for diagnostic and prognostic purposes;and (3) staging/prediction models that predict clinically relevant parameters such as tumorsize, pleural invasion, and TNM stage. We rigorously examined 50 peer-reviewed andpreprint research works, reporting shared datasets, model structures, performance measures(e.g., Dice, accuracy, AUC), strengths and weaknesses. Our discussion points out keychallenges—such as limited annotated data, absence of multi-center external validation, andexplainability shortcomings—and suggests future directions like multi-modal fusion,federated learning, and prospective clinical validation. Overall, this survey provides researchers and practitioners with a systematic overview of the state-of-the-art, challenges to fill in, and real-world considerations on how to construct robust deep learning systems for early-stage lung cancer staging.
