The Evolution of Deep Learning in Lung Cancer Prediction: a Review

Lung cancer remains the leading cause of cancerrelated mortality worldwide, underscoring the critical need for early and accurate diagnosis. The period from 2015 to 2025 witnessed a significant transformation in computational approaches for lung cancer prediction, marked by a rapid shift from traditional machine learning to sophisticated deep learning models. This comprehensive review systematically examines this evolution, analyzing 42 research papers to catalog the datasets, preprocessing techniques, feature extraction methodologies, and deep learning architectures that have defined the state of the art. Our analysis reveals that convolutional neural networks (CNNs), particularly 3D architectures and ensemble models, have become predominant, demonstrating a substantial improvement in performance. The findings indicate that these deep learning models achieve accuracy rates between 87 % and 95 % on benchmark datasets like LIDC-IDRI, significantly outperforming traditional methods in sensitivity, specificity, and area under the curve (AUC) metrics. This review synthesizes these advancements, discusses persistent challenges such as data scarcity and model interpretability, and highlights the promising yet complex path toward clinical integration.